DETAILED ACTION
This action is responsive to the claims filed on 03/31/2024. Claims 1-21 are pending for examination.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 07/20/2025 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-21 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Statutory Categories
Claims 1-19 are directed to a method.
Claim 20 is directed to a system.
Claim 21 is directed to a computer-readable medium.
Independent Claims – Claims 1, 20, and 21
Step 2A Prong 1: Does the claim recite an abstract idea, law of nature, or natural phenomenon?
Yes. Independent claims 1, 20, and 21 recites limitations that are abstract ideas in the form of mental processes:
Claim 1 recites:
… performing … an analysis of said REP, and …generating an … output indicating at least: a proposed minimal set of elevated permissions that are estimated …to be needed and also sufficient for achieving a task described in the REP (performing an analysis, and generated an output based on an LLM, stated at a high level of generality, is being considered as a mental process of evaluation which can reasonably be performed by human mind or with aid of pen and paper)
Claim 1 also recites the following additional elements for the purposes of Step 2A Prong Two
analysis:
A computerized method for automatically evaluating requests for escalation of privileges in an organization, the computerized method comprising: (a) receiving a Request for Escalation of Privileges (REP), in a natural language, that indicates a request from a Requesting User to elevate access privileges to a computerized organizational resource; (b) automatically feeding the REP as input into a fine-tuned Large Language Model (LLM); (receiving and inputting data is merely data gathering and is considered insignificant extra-solution activity under MPEP 2106.05 (g))
automatically, … by said fine-tuned LLM … an LLM-based …by the LLM…” should be analyzed as additional elements of mere (this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
The additional limitations fail step 2A Prong 2 of the 101 analysis because they do not transform the claim into a practical application. These limitations are too abstract or lack technical improvement that would make the concept practically useful. Without clear utility or integration into a specific field, the claim does not relate to any particular application. It does not meet the requirements of Step 2A Prong 2, as it fails to make the concept meaningfully applicable in practice.
Since the claim as a whole, looking at the additional elements individually and in combination, does not contain any other additional elements that are indicative of integration into a practical application, the claim is “directed” to an abstract idea.
This claim recites the following additional elements for the purposes of Step 2B analysis:
A computerized method for automatically evaluating requests for escalation of privileges in an organization, the computerized method comprising: (a) receiving a Request for Escalation of Privileges (REP), in a natural language, that indicates a request from a Requesting User to elevate access privileges to a computerized organizational resource; (b) automatically feeding the REP as input into a fine-tuned Large Language Model (LLM); (receiving and inputting data is merely data gathering and is considered insignificant extra-solution activity under MPEP 2106.05 (g), furthermore under step 2B it should be noted that the courts have recognized receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information) as well-understood, routine, and conventional activity.)
automatically, … by said fine-tuned LLM … an LLM-based …by the LLM…” should be analyzed as additional elements of mere (this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
The claim also fails Step 2B of the analysis because the additional limitations do not amount to significantly more than the abstract idea itself. The additional limitations do not enhance the claim in a way that would move it beyond its abstract ideas as they minimally elaborate on the core concept without adding any inventive or technical substance. Considering the additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Claims 20 and 21 recites limitations substantially similar to claim 1, as such a similar analysis applies.
Claim 20 recites the following additional limitation for consideration:
A system comprising: one or more hardware processors, that are configured to execute code, and that are operably associated with one or more memory units that are configured to store code; wherein the one or more hardware processors are configured (computer components recited at a high level of generality such that they are merely invoking computers as a tool to perform the judicial exception, see MPEP 2106.05(f))
Claim 20 recites the following additional limitation for consideration:
A non-transitory storage medium having stored thereon instructions that, when executed by a machine, cause the machine to perform (computer components recited at a high level of generality such that they are merely invoking computers as a tool to perform the judicial exception, see MPEP 2106.05(f))
Dependents of Claim 1
The remaining dependent claims corresponding to independent claim 1 do not recite additional elements, whether considered individually or in combination, that are sufficient to integrate the judicial exception into a practical application or amount to significantly more than the judicial exception. The analysis of which is shown below:
The claims below recite additional limitations which fail step 2A Prong 2 of the 101 analysis because they do not transform the claim into a practical application. These limitations are too abstract or lack technical improvement that would make the concept practically useful. Without clear utility or integration into a specific field, the claim does not relate to any particular application. It does not meet the requirements of Step 2A Prong 2, as it fails to make the concept meaningfully applicable in practice.
The claims also fails Step 2B of the analysis because the additional limitations do not amount to significantly more than the abstract idea itself. The additional limitations do not enhance the claim in a way that would move it beyond its abstract ideas as they minimally elaborate on the core concept without adding any inventive or technical substance. The claims are unpatentable.
Claim 2 recites the additional limitation of:
The computerized method of claim 1, wherein step (c) further comprises: generating said… output that further comprises a textual reasoning for inclusion of at least one elevated permission in said proposed minimal set of elevated permissions. (generating an output comprising a textual reasoning, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
automatically generating said LLM-based output (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 3 recites the additional limitation of:
The computerized method of claim 1, wherein step (c) further comprises: generating said… output that comprises at least one elevated permission that was not explicitly requested by said Requesting User in said REP, and that… predicts to be necessary for achieving said task described in said REP. (generating an LLM-based output, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
automatically generating said LLM-based output… and the fine-tuned LLM predicts (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 4 recites the additional limitation of:
The computerized method of claim 1, wherein step (c) further comprises: generating said… output that comprises machine-readable code that, when executed by a machine, causes automatic granting of said elevated permissions to said Requesting User. (generating an LLM-based output, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
automatically generating said LLM-based output (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 5 recites the additional limitation of:
The computerized method of claim 4, wherein step (c) further comprises: generating said… output that comprises machine-readable code that, when executed by said machine, causes automatic revocation of said elevated permissions after a pre-defined time-period elapses. (generating an LLM-based output, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
automatically generating said LLM-based output (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 6 recites the additional limitation of:
The computerized method of claim 1, wherein step (c) further comprises: generating said… output that further comprises: a textual LLM-generated explanation of one or more adverse consequences that the… estimates to occur if said proposed minimal set of elevated permissions is not approved. (generating an LLM-based output, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
automatically generating said LLM-based output… the fine-tuned LLM estimates (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 7 recites the additional limitation of:
The computerized method of claim 1, wherein step (c) further comprises: generating said… output that further comprises: a textual LLM-generated explanation of an estimated level of urgency that the… estimates to be associated with said task described in said REP. (generating an LLM-based output, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
automatically generating said LLM-based output… the fine-tuned LLM estimates (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 8 recites the additional limitation of:
The computerized method of claim 1, wherein step (c) further comprises: generating said… output that comprises definitions for automatic creation of a Custom Role that includes exactly said proposed minimal set of elevated permissions and no additional permissions. (generating an LLM-based output, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
automatically generating said LLM-based output… (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 9 recites the additional limitation of:
The computerized method of claim 1, wherein step (c) further comprises: generating said… output that comprises definitions for automatic assignment of one or more pre-defined roles that together include exactly said proposed minimal set of elevated permissions and no additional permissions. (generating an LLM-based output, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
automatically generating said LLM-based output… (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 10 recites the additional limitation of:
The computerized method of claim 1, wherein step (b) further comprises: feeding to said fine-tune LLM a signal indicating whether Custom Roles are supported or only Pre-Defined Roles are supported; (Under step 2A prong 2: receiving and inputting data is merely data gathering and is considered insignificant extra-solution activity under MPEP 2106.05 (g), furthermore under step 2B it should be noted that the courts have recognized receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information) as well-understood, routine, and conventional activity.)
if Custom Roles are supported, then the method comprises: generating said… output that comprises definitions for automatic creation of a Custom Role that includes exactly said proposed minimal set of elevated permissions and no additional permissions; (generating an LLM-based output, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
if Custom Roles are not supported and only Pre-Defined Roles are supported, then the method comprises: generating said… output that comprises definitions for automatic assignment of one or more pre-defined roles that together include exactly said proposed minimal set of elevated permissions and no additional permissions. (generating an LLM-based output, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
automatically generating said LLM-based output… (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 11 recites the additional limitation of:
The computerized method of claim 1, wherein step (b), of automatically feeding the REP as input into the fine-tuned LLM, comprises: (b1) enriching said REP by performing Retrieval Augmented Generation (RAG), (Under step 2A prong 2 and step 2B: this limitation is merely applying the use of Retrieval Augmented Generation at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
and generating a RAG-enriched prompt; (generating an LLM-based output, stated at a high level of implementation is being considered as mental processes of evaluation which can reasonably be performed in human mind with aid of pen and paper)
(b2) feeding said RAG-enriched prompt as input to the fine-tuned LLM. (Under step 2A prong 2: receiving and inputting data is merely data gathering and is considered insignificant extra-solution activity under MPEP 2106.05 (g), furthermore under step 2B it should be noted that the courts have recognized receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information) as well-understood, routine, and conventional activity.)
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 12 recites the additional limitation of:
The computerized method of claim 11, wherein enriching said REP is performed using a RAG unit that has access to a vector database that was constructed with embeddings from a plurality of documentation items corresponding to a plurality of computerized services and cloud-based services. (Under step 2A prong 2 and step 2B: this limitation is merely applying the use of Retrieval Augmented Generation at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 13 recites the additional limitation of:
The computerized method of claim 12, wherein enriching said REP comprises: adding to said REP a context that includes selected segments of said documentation items that said… units determines to have semantic similarity to said REP beyond a pre-defined threshold level of semantic similarity. (Adding a context to a Request for Escalation of Privileges for context that a certain similarity is above a threshold is being considered a mental process of evaluation which can reasonably be performed in human mind or with aid of pen and paper)
RAG units (this limitation is merely directed to a field of use (RAG units) see MPEP 2106.05(h))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 14 recites the additional limitation of:
The computerized method of claim 12, wherein enriching said REP is performed using a RAG unit that has access to a vector database that was constructed with embeddings from data-sources describing roles and permissions in said organization. (Under step 2A prong 2 and step 2B: this limitation is merely applying the use of Retrieval Augmented Generation at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 15 recites the additional limitation of:
The computerized method of claim 1, further comprising: constructing said fine-tuned LLM by fine-tuning an already-existing LLM with an annotated dataset that includes: (i) textual natural-language descriptions of tasks that users would like to achieve, and (b) for each of said tasks, a list of minimal permissions that are required for completing said task. (fine-tuning a data set, which paragraph [0030] of the specification comprises as “modification of weights and biases” with additional annotated data is being considered a mental process of evaluation which can reasonably be performed in human mind or with aid of pen and paper)
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 16 recites the additional limitation of:
The computerized method of claim 1, further comprising: (d) automatically revoking said elevated permissions, after elapsing of a pre-defined time-period that was indicated in said output of said fine-tuned LLM in response to said REP. (revoking permissions in response to a predetermined time period elapsing is being considered a mental process of evaluation which can reasonably be performed in human mind or with aid of pen and paper)
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 17 recites the additional limitation of:
The computerized method of claim 1, further comprising: (d1) automatically providing the output of said fine-tuned LLM, in response to said REP, to a computing device of a Reviewing User; (d2) receiving from said Reviewing User either an approval or a rejection of the proposal of minimal set of elevated permissions that was generated by said fine-tuned LLM; (Under step 2A prong 2: receiving and inputting data is merely data gathering and is considered insignificant extra-solution activity under MPEP 2106.05 (g), furthermore under step 2B it should be noted that the courts have recognized receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information) as well-understood, routine, and conventional activity.)
(d3) if in step (d2) an approval signal was received from the Reviewing User, then: invoking the minimal set of elevated permissions that were indicated in the proposal generated, (invoking permissions in response to a predetermined approval signal being considered a mental process of evaluation which can reasonably be performed in human mind or with aid of pen and paper)
and automatically sending a message to the Requesting User indicating which permissions were elevated; (d4) if in step (d2) a rejection signal was received from the Reviewing User, then: automatically sending a message to the Requesting User indicating that the REP was rejected. (Under step 2A prong 2: receiving and inputting data is merely data gathering and is considered insignificant extra-solution activity under MPEP 2106.05 (g), furthermore under step 2B it should be noted that the courts have recognized receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information) as well-understood, routine, and conventional activity.)
automatically… by said fine-tuned LLM (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the automatic use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 18 recites the additional limitation of:
The computerized method of claim 1, further comprising: (d1) automatically providing the output of said fine-tuned LLM, in response to said REP, to a computing device of a Reviewing User; (d2) receiving from said Reviewing User one or more clarification questions, and feeding the one or more clarification questions based into said fine-tuned LLM; (Under step 2A prong 2: receiving and inputting data is merely data gathering and is considered insignificant extra-solution activity under MPEP 2106.05 (g), furthermore under step 2B it should be noted that the courts have recognized receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information) as well-understood, routine, and conventional activity.)
(d3) in response to the one or more clarification questions from the Reviewing User, generating… at least one of: (i) one or more responses to said one or more clarification questions, (ii) a modified output of minimal set of elevated permissions. (generating responses to users questions is being considered a mental process of evaluation which can reasonably be performed in human mind or with aid of pen and paper)
by said fine-tuned LLM (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim 19 recites the additional limitation of:
The computerized method of claim 1, wherein the REP from the Requesting User is in a first natural language, and is processed, (this limitation is merely directing the REP into a field of use (natural language) see MPEP 2106.05(h))
and wherein output from said fine-tuned LLM is provided to the Reviewing User in a second, different, natural language. (Under step 2A prong 2: receiving and inputting data is merely data gathering and is considered insignificant extra-solution activity under MPEP 2106.05 (g), furthermore under step 2B it should be noted that the courts have recognized receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information) as well-understood, routine, and conventional activity.)
by the fine-tuned LLM (For the purposes of Step 2A prong 2 and Step 2B: this limitation is merely applying the use of an LLM at a high level of generality such that is merely applying the exception, see MPEP 2106.05(f))
Since the claim does not recite additional elements that either integrate the judicial exception into a practical application, nor provide significantly more than the judicial exception, the claim is not patent eligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or non-obviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 3-8, 15-16 and 20-21 are rejected under 35 U.S.C. 103 as being unpatentable over Ailawadi et al. (US 2024/0330421 A1), hereafter referred to as Ailawadi, in view of Zaman et al., (US 2024/0372871 A1), hereafter referred to as Zaman.
Claim 1: Ailwadi teaches the following:
A computerized method for automatically evaluating requests for escalation of privileges in an organization, (Ailawadi, paragraph 27, “The fulfillment of requests in a timely manner and with accuracy is very important to keep organizations safe and secure, hence the self service requests module. This takes inputs from the end user and, based upon the role based permissions of the user, it raises requests to the relevant authorities along with serving the parts of request in real time which do not need approval.”, Ailawadi discloses a computerized enterprise self-service module that receives user requests, evaluates the requests in view of the user’s role-based permissions, raises requests requiring approval to the appropriate authorities, and automatically serves portions that do not require approval. The disclosed processing therefore teaches automatically evaluating organizational requests concerning access or privilege escalation.)
the computerized method comprising: (a) receiving a Request for Escalation of Privileges (REP), in a natural language, that indicates a request from a Requesting User to elevate access privileges to a computerized organizational resource; (Ailawadi, paragraph 28, “Self-service addresses requests to the system for services provided by the enterprise to its employees. For example, self-service in invoked when an employee wants to request a badge, when somebody needs access, when they are traveling to a different site, when they want to update their emergency contact information, etc.”, Ailawadi expressly receives enterprise employee requests in which an employee “needs access” or requests a badge. The employee is the requesting user, and the request seeks access to an organizational resource. Such a request corresponds to the claimed REP.
Paragraph 29, “Embodiments of the invention use natural language processing (NLP). The user can ask any questions, e.g. they can ask what they need and the system immediately returns a responsive page to the user rather than requiring the user to click several different links and submit additional information. Thus, a page is returned to the user who then merely has to hit one button, review, and submit. This feature of the invention saves the user’s time, saves the user’s clicks and, based on natural language, allows the user to ask what he needs and the system then gets it in front of him.”, Ailawadi expressly teaches that the employee states what is needed in natural language and that the system processes the natural-language request to provide the appropriate responsive page. This teaches receiving the REP in natural language.)
(b) automatically feeding the REP as input into a fine-tuned Large Language Model (LLM); (Ailawadi, paragraph 83, “FIG. 5 is a block diagram that shows an adaptive language model-based solution for interactive security and safety with data privacy according to an embodiment of the invention. The end user can access the user interface 50 and provide inputs in the form of natural language text, voice, or image, e.g. via a search bar, submit button, a display areas, etc. The inputs taken from the user is processed by a preprocessing block 51, which in embodiments of the invention may be an internal application which can operate on premises or over the cloud based upon user preference. The preprocessing block interacts with a module 57 that passes inputs and prompts to a large language model 56. The preprocessing block also interacts with a JSON generator 58 to receive responses from the large language model.”, The employee’s natural-language access request is an end-user input. Ailawadi’s preprocessing block automatically passes that input and associated prompts to the LLM, thereby teaching the claimed feeding step.)
(c) automatically performing by said fine-tuned LLM an analysis of said REP, and automatically generating an LLM-based output (Ailawadi, paragraph 84, “Embodiments of the invention leverage large language models to instruct a software program or system which is meant to be invoked by a particular syntax, e.g. software is invoked by a function or method, e.g. an SQL database is invoked by an SQL query. In an input to the LLM the user provides a list of functions/sql queries/syntax along with the text in the form of prompts to support the LLM model to adapt to the proprietary information and syntax. The LLM model, which is fine tuned with the information provided, as explained below, then generates an output in the form of components of the function/query/syntax, e.g. name of function, attributes in the function and any other values which are relevant to invoke the functionality that function is responsible for in the intended software.”, Ailawadi expressly identifies the model as fine-tuned. The model interprets the text and prompt information and generates structured function, query, or syntax components, including function names, attributes, and values. Those generated components are an LLM-based output resulting from analysis of the request.)
Zaman, in the same field of machine learning, teaches the following which Ailawadi fails to teach:
an LLM-based output indicating at least: a proposed minimal set of elevated permissions that are estimated by the LLM to be needed and also sufficient for achieving a task described in the REP. (Zaman, paragraph 26, “As shown by reference number 110, the role generation device may determine or generate one or more roles (e.g., one or more IAM roles) for the task based on the deployment pattern. For example, the one or more roles may be least privilege roles. For example, based on the information associated with the deployment pattern, the role generation device may determine what resources and/or services (e.g., associated with the cloud computing engine) are needed to perform the task. The role generation device may generate a role (e.g., an IAM role) for the task that only allows access to the resources and/or services that are needed to perform the task.”, Zaman determines the resources and services needed to perform the task and generates a least-privilege role permitting access only to those needed resources and services. The role therefore defines a minimal permission set that is sufficient to perform the task while excluding permissions that are not needed.)
Ailawadi teaches a natural-language, fine-tuned-LLM enterprise system that receives an employee access request and generates structured output for executing enterprise functions, but does not expressly require the resulting access to be limited to the minimum permissions sufficient for the described task. Zaman teaches determining the resources and services needed for a task and generating a least-privilege IAM role limited to those resources and services. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to incorporate Zaman’s task-based least-privilege role generation into Ailawadi’s natural-language access-request workflow to provide sufficient task access while reducing excess privilege and the associated security risk. The combination would have used each reference’s known functionality for its established purpose and would have yielded the predictable result of a natural-language access system that proposes only the permissions needed to complete the described task
Claim 3: Ailawadi and Zaman teaches the method of claim 1, Zaman further teaches:
The computerized method of claim 1, wherein step (c) further comprises: automatically generating said LLM-based output that comprises at least one elevated permission that was not explicitly requested by said Requesting User in said REP, and that the fine-tuned LLM predicts to be necessary for achieving said task described in said REP. (Zaman, paragraph 27, “For example, the role generation device may analyze the deployment pattern to determine the resources and/or services (e.g., associated with the cloud computing engine) that are needed to perform the task. In some implementations, the role generation device may analyze one or more previous task occurrences to identify resources and/or services (e.g., associated with the cloud computing engine) that were actually used to perform the task during the one or more previous task occurrences. The role generation device may create a role (e.g., an IAM role) that grants permissions to access the resources and/or services (e.g., associated with the cloud computing engine) that were actually used to perform the task during the one or more previous task occurrences.”, Zaman derives required permissions from resources and services actually used during prior occurrences of the task rather than limiting the role to permissions expressly named by a requester. Accordingly, where the user describes the task but omits a required permission, Zaman’s analysis identifies and includes the omitted permission as necessary to perform the task.)
Claim 4: Ailawadi and Zaman teaches the method of claim 1, Ailawadi further teaches:
The computerized method of claim 1, wherein step c) further comprises: automatically generating said LLM-based output that comprises machine-readable code (Ailawadi, paragraph 84, “Embodiments of the invention leverage large language models to instruct a software program or system which is meant to be invoked by a particular syntax, e.g. software is invoked by a function or method, e.g. an SQL database is invoked by an SQL query. In an input to the LLM the user provides a list of functions/sql queries/syntax along with the text in the form of prompts to support the LLM model to adapt to the proprietary information and syntax. The LLM model, which is fine tuned with the information provided, as explained below, then generates an output in the form of components of the function/query/syntax, e.g. name of function, attributes in the function and any other values which are relevant to invoke the functionality that function is responsible for in the intended software.”, Ailawadi’s generated function, query, or syntax components include a function name, attributes, and values used to invoke software functionality. Those structured components constitute machine-readable code or machine readable instructions.)
that, when executed by a machine, causes automatic granting of said elevated permissions to said Requesting User. (Ailawadi, paragraph 87, “The information received by the execution block is used to trigger the respective function and its dependencies. The function could be responsible for a functionality such as, e.g. “block access of an identity for four hours,” “grant access of a particular access level to the mentioned badge id,” “show how many people are present in the facility right now,” “who all are accessed VPN last night,” and so on.”, Ailawadi executes the generated function information and expressly identifies granting a particular access level to an identified badge as an executable function. When the generated function implements Zaman’s proposed least privilege role, execution automatically grants the proposed elevated permissions to the requesting user.)
Claim 5: Ailawadi and Zaman teaches the method of claim 1, Ailawadi further teaches:
The computerized method of claim 4, wherein step (c) further comprises: automatically generating said LLM-based output that comprises machine-readable code (Ailawadi, paragraphs 84 and 87-88, “The LLM model, which is fine tuned with the information provided, as explained below, then generates an output in the form of components of the function/query/syntax, e.g. name of function, attributes in the function and any other values which are relevant to invoke the functionality that function is responsible for in the intended software. The information received by the execution block is used to trigger the respective function and its dependencies. The function could be responsible for a functionality such as, e.g. “block access of an identity for four hours.””, Ailawadi generates executable function syntax and identifies blocking an identity’s access as an exemplary function. Blocking previously granted access constitutes revocation of the corresponding elevated permission.)
Zaman further teaches:
that, when executed by said machine, causes automatic revocation of said elevated permissions after a pre-defined time-period elapses. (Zaman, paragraph 34, “In some implementations, the role generation device may cause the one or more roles associated with the task to be stored in the data structure for an amount of time or a duration. For example, the role generation device may generate the one or more roles to be associated with the duration. The duration may indicate an amount of time for which the one or more roles are valid. For example, the role generation device may delete or remove the one or more roles (e.g., the one or more least privilege IAM roles) from the data structure after an amount of time (e.g., an amount of time from the time at which the one or more roles were initially stored in the data structure). For example, the amount of time may be the amount of time for which the one or more roles are valid.”, Zaman associates the least-privilege role with a validity period and deletes or removes the role after that period. Removing the role terminates its permissions and therefore teaches automatically revoking the permissions after a predefined time period elapses.)
Claim 6: Ailawadi and Zaman teaches the method of claim 1, Zaman further teaches:
The computerized method of claim 1, wherein step (c) further comprises: automatically generating said LLM-based output that further comprises: a textual LLM-generated explanation of one or more adverse consequences that the fine-tuned LLM estimates to occur if said proposed minimal set of elevated permissions is not approved. (Zaman, paragraph 36, “Additionally, causing the one or more roles to be associated with a limited duration improves an accuracy of the resources and/or services that are granted via the one or more roles. For example, at a first time, one or more resources or services may be needed to perform the task. However, at a second (e.g., later) time, an additional resource or service may be needed to perform the task. If the one or more roles (e.g., that grant access to the resource or service) are indefinitely stored in the data structure, then an occurrence of the task at, or after, the second time may not grant access to the additional resource or service that is actually needed to perform the task, which may result in the task not being performed and/or only partially being performed. This may consume processing resources, computing resources, and/or network resources attempting to perform the task using a role (e.g., an IAM role) that does not grant access to all resources or services needed to perform the task.”, Zaman expressly identifies adverse consequences of not providing the required permission set: the task may not be performed, may be only partially performed, and may waste processing, computing, and network resources.)
Claim 7: Ailawadi and Zaman teaches the method of claim 1, Ailawadi further teaches:
The computerized method of claim 1, wherein step (c) further comprises: automatically generating said LLM-based output that further comprises: a textual LLM-generated explanation of an estimated level of urgency that the fine-tuned LLM estimates to be associated with said task described in said REP. (Ailawadi, paragraph 37, “A playbook provides a recommendation that is made when data stored in the enterprise is useful to address an immediate need or concern. For example, a user such as a security guard may query: “I see a suspicious package in the parking lot. What should I do? What is the protocol of the enterprise for this situation?””, Ailawadi uses the LLM-based recommendation system to address an “immediate need or concern” and provides natural-language guidance responsive to that circumstance. Characterizing the request as immediate and explaining the appropriate response teaches an LLM-generated explanation reflecting the urgency associated with the task.)
Claim 8: Ailawadi and Zaman teaches the method of claim 1, Ailawadi further teaches:
The computerized method of claim 1, wherein step (c) further comprises: automatically generating said LLM-based output that comprises definitions for automatic creation of a Custom Role (Ailawadi, paragraph 81, “Generative Model Training. Generative model, e.g. Generative Adversarial Network or Variational Autoencoder, learns the underlying patterns in the data and generates new configurations of access levels and job roles leveraging the similarity between the access levels and job roles respectively as the basis. The model is fine-tuned guided with the samples of the ideal desired output for better accuracy.”, Ailawadi generates new configurations of access levels and job roles rather than merely selecting an existing role. Such a generated configuration defines a custom role, and Ailawadi’s structured-output and execution architecture makes the generated definition available for automatic implementation.)
Zaman further teaches:
that includes exactly said proposed minimal set of elevated permissions and no additional permissions. (Zaman, paragraph 26, “As shown by reference number 110, the role generation device may determine or generate one or more roles (e.g., one or more IAM roles) for the task based on the deployment pattern. For example, the one or more roles may be least privilege roles. For example, based on the information associated with the deployment pattern, the role generation device may determine what resources and/or services (e.g., associated with the cloud computing engine) are needed to perform the task. The role generation device may generate a role (e.g., an IAM role) for the task that only allows access to the resources and/or services that are needed to perform the task.”, A role that permits access only to the resources and services needed to perform the task contains the task’s minimal permission set and excludes additional permissions.)
Claim 15: Ailawadi and Zaman teaches the method of claim 1, Ailawadi further teaches:
The computerized method of claim 1, further comprising: constructing said fine-tuned LLM by fine-tuning an already-existing LLM (Ailwadi, paragraphs 62-63, “Fine-tuning is the practice of modifying an existing, pretrained language model by training it in a supervised fashion on a specific task, e.g. sentiment analysis, named entity recognition, or part-of-speech tagging. It is a form of transfer learning, i.e. a technique in machine learning (ML) in which knowledge learned from a task is re-used to boost performance on a related task… Fine tuning a large language model involves adjusting the parameters or weights of a pre-trained model to adapt it to a new task.”, Ailawadi expressly starts with an existing pretrained language model and modifies its parameters or weights through supervised fine-tuning for a specific task.)
with an annotated dataset that includes: (i) textual natural-language descriptions of tasks that users would like to achieve, (Ailawadi, paragraph 93, “Inputs and prompts to the large model are generated when preprocessing block performs hard fine tuning 53 on a syntactical data example SQL queries. As shown in the examples above, the text and corresponding SQL queries are fed into the training data through which the AI models learn various table names, column names, and their relations with each other to generate an SQL query on a text which could be very new for it but is intended for same table, column, and syntax structure.”, Ailawadi’s training data pairs natural-language text with corresponding desired structured outputs. Those paired examples constitute an annotated dataset of textual descriptions and target outputs.)
Zaman further teaches:
and (b) for each of said tasks, a list of minimal permissions that are required for completing said task. (Zaman, paragraphs 27-28, “The role generation device may analyze one or more previous task occurrences to identify resources and/or services (e.g., associated with the cloud computing engine) that were actually used to perform the task during the one or more previous task occurrences. The role generation device may create a role (e.g., an IAM role) that grants permissions to access the resources and/or services (e.g., associated with the cloud computing engine) that were actually used to perform the task during the one or more previous task occurrences. In other words, the role generation device may generate or create a role (e.g., an IAM role) that indicates one or more security permissions. The one or more security permissions may be associated with controlling access to one or more resources and/or one or more services associated with the cloud computing environment for performing the task.”, Zaman associates each task with a least-privilege role listing the permissions derived from the resources and services actually needed to perform that task. Using the natural-language task description as a training input and Zaman’s associated permission list as the desired output yields the claimed annotated training examples.)
Claim 16: Ailawadi and Zaman teaches the method of claim 1, Zaman further teaches:
The computerized method of claim 1, further comprising: (d) automatically revoking said elevated permissions, (Zaman, paragraph 34, “In some implementations, the role generation device may cause the one or more roles associated with the task to be stored in the data structure for an amount of time or a duration. For example, the role generation device may generate the one or more roles to be associated with the duration. The duration may indicate an amount of time for which the one or more roles are valid. For example, the role generation device may delete or remove the one or more roles (e.g., the one or more least privilege IAM roles) from the data structure after an amount of time (e.g., an amount of time from the time at which the one or more roles were initially stored in the data structure). For example, the amount of time may be the amount of time for which the one or more roles are valid.”, Zaman automatically removes the least-privilege role after its validity duration. Because the role supplies the elevated permissions, removal of the role automatically revokes those permissions after the predefined period elapses.)
Ailawadi further teaches:
after elapsing of a pre-defined time-period that was indicated in said output of said fine-tuned LLM in response to said REP. (Ailawadi, paragraph 33, “‘I’m travelling to building A and I need access to it for a week.’”; Ailwadi, paragraph 84, ““The LLM model, which is fine tuned with the information provided, as explained below, then generates an output in the form of components of the function/query/syntax, e.g. name of function, attributes in the function and any other values which are relevant to invoke the functionality that function is responsible for in the intended software.”, Ailawadi expressly teaches that the natural-language access request includes a predefined access period e.g. “for a week”, and that the fine-tuned LLM generates function output containing the attributes and other values relevant to invoking the requested functionality. For the disclosed request for access lasting one week, the one-week duration is a value relevant to invoking the requested time-limited access function and is therefore indicated in the LLM-generated output. Ailawadi is relied upon for teaching that the predefined duration from the REP is indicated as a value in the LLM output.)
Claims 20 and 21 recite limitations substantially similar to claim 1, as such a similar analysis applies.
Claim 20 recites the following additional limitation for consideration which Ailawadi further teaches:
A system comprising: one or more hardware processors, that are configured to execute code, and that are operably associated with one or more memory units that are configured to store code; wherein the one or more hardware processors are configured to perform a computerized process for automatically evaluating requests for escalation of privileges in an organization, (Ailawadi, paragraph 143, “The computing system 60 may include one or more central processing units (‘processors’) 62, memory 61, input/output devices 64, e.g. keyboard and pointing devices, touch devices, display devices, storage devices 66, e.g. disk drives, and network adapters 330, e.g. network interfaces, that are connected to an interconnect 65. The interconnect 65 is illustrated as an abstraction that represents any one or more separate physical buses, point to point connections, or both connected by appropriate bridges, adapters, or controllers.”; Paragraph 28, “Self-service addresses requests to the system for services provided by the enterprise to its employees. For example, self-service in invoked when an employee wants to request a badge, when somebody needs access, when they are traveling to a different site, when they want to update their emergency contact information, etc.”, Ailawadi discloses a computing system comprising one or more central processing units. Ailawadi expressly states that the memory and storage devices are computer-readable storage media, including non-transitory media, that store software or firmware instructions used to program processors 62 to perform the disclosed computerized operations. Ailawadi further discloses an organizational self-service request module that receives input from an end user, evaluates the request based on the user’s existing role-based permissions, and either processes the request or raises it to an appropriate authority. Accordingly, Ailawadi teaches a system having hardware processors configured to execute code stored in operably associated memory units to perform a computerized process for automatically evaluating organizational requests for escalation or elevation of access privileges.)
Claim 21 recites the following additional limitation for consideration which Ailawadi further teaches:
A non-transitory storage medium having stored thereon instructions that, when executed by a machine, cause the machine to perform a method for automatically evaluating requests for escalation of privileges in an organization(Ailawadi, paragraph 143, “The computing system 60 may include one or more central processing units (‘processors’) 62, memory 61, input/output devices 64, e.g. keyboard and pointing devices, touch devices, display devices, storage devices 66, e.g. disk drives, and network adapters 330, e.g. network interfaces, that are connected to an interconnect 65. The interconnect 65 is illustrated as an abstraction that represents any one or more separate physical buses, point to point connections, or both connected by appropriate bridges, adapters, or controllers.”; Paragraph 28, “Self-service addresses requests to the system for services provided by the enterprise to its employees. For example, self-service in invoked when an employee wants to request a badge, when somebody needs access, when they are traveling to a different site, when they want to update their emergency contact information, etc.”, Ailawadi discloses a computing system comprising one or more central processing units. Ailawadi expressly states that the memory and storage devices are computer-readable storage media, including non-transitory media, that store software or firmware instructions used to program processors 62 to perform the disclosed computerized operations. Ailawadi further discloses an organizational self-service request module that receives input from an end user, evaluates the request based on the user’s existing role-based permissions, and either processes the request or raises it to an appropriate authority. Accordingly, Ailawadi teaches a system having hardware processors configured to execute code stored in operably associated memory units to perform a computerized process for automatically evaluating organizational requests for escalation or elevation of access privileges.)
Claims 2, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Ailawadi, in view of Zaman and Badawy et al., (US 2023/0035335 A1), hereafter referred to as Badawy.
Claim 2: Ailawadi and Zaman teaches the method of claim 1, Badawy, in the same field of machine learning, further teaches the following which the above prior art fails to teach:
The computerized method of claim 1, wherein step (c) further comprises: automatically generating said LLM-based output that further comprises a textual reasoning for inclusion of at least one elevated permission in said proposed minimal set of elevated permissions. (Badawy, paragraph 94, “To make the recommendation system more usable, the system can provide interpretations of the recommendations to users. For example, the peer ids, max similarity, output from ML model interpretation layer and other stats are generated as part of the output to help explain why a particular item was recommended. The user interface (UI) can feature a method to allow users to request such explanations to help with the approach and ML model’s transparency. As another example, if a user requests more information about a recommended entitlement, the system can provide detailed information about why access to a particular entitlement was recommended. Two examples of model interpretability tools which can be used for this purpose are LIME (local interpretable model-agnostic explanations) and SHAP (SHapley Additive exPlanations). Either interpretability tool can be used to interpret what features contributed to the generated recommendation, as one skilled in the art would understand. This information may help a user decide whether or not to request a recommended access item. This information may also help a manager or other decision maker in the approval process.”, Badawy expressly provides output explaining why a particular entitlement was recommended and what features contributed to the recommendation. Because an entitlement is an access permission, the disclosed interpretation is textual reasoning supporting inclusion of the permission in the proposed set. Any arbitrary LLM output provides the known natural-language mechanism for presenting Badawy’s explanation.)
Badawy is directed to automated access recommendations and teaches providing explanations of why an entitlement was recommended so that a user, manager, or other decision-maker can evaluate the recommendation. It would have been obvious to one of ordinary skill in the art before the effective filing date of the invention to provide Badawy’s recommendation explanation with the Ailawadi-Zaman permission proposal because doing so would improve transparency, auditability, and the reviewer’s ability to make an informed approval decision. Presenting that known explanation through Ailawadi’s existing natural language interface would have been a predictable use of the interface’s textual-output capability.
Claim 18: Ailawadi and Zaman teaches the method of claim 1, Ailawadi further teaches:
and feeding the one or more clarification questions based into said fine-tuned LLM; (Ailawadi, paragraph 83-84, “The end user can access the user interface 50 and provide inputs in the form of natural language text, voice, or image, e.g. via a search bar, submit button, a display areas, etc. The preprocessing block interacts with a module 57 that passes inputs and prompts to a large language model 56. The LLM model, which is fine tuned with the information provided, as explained below, then generates an output in the form of components of the function/query/syntax, e.g. name of function, attributes in the function and any other values which are relevant to invoke the functionality that function is responsible for in the intended software.”, Ailawadi accepts subsequent natural-language user input through the interface, passes the input and prompts to the fine-tuned LLM, and receives an LLM-generated output. Using that same disclosed interaction for the reviewer’s clarification question teaches feeding the question into the LLM and generating a response.)
Zaman further teaches:
(d3) in response to the one or more clarification questions from the Reviewing User, generating by said fine-tuned LLM at least one of: (i) one or more responses to said one or more clarification questions, (ii) a modified output of minimal set of elevated permissions. (Zaman, claim 11, “Receiving an indication of an update associated with a least privilege IAM role, from the one or more least privilege IAM roles, indicating one or more updated security permissions associated with the least privilege IAM role; and updating, in the data structure, information associated with the least privilege IAM role to indicate the one or more updated security permissions.”, Zaman expressly receives updated permission information for a least-privilege role and modifies the stored role to reflect the updated permissions. When the reviewer’s clarification changes the understood task requirements, applying Zaman’s disclosed update produces a modified minimal permission-set output.)
Badawy, in the same field of machine learning, teaches the following which the above prior art fails to teach:
The computerized method of claim 1, further comprising: (d1) automatically providing the output of said fine-tuned LLM, in response to said REP, to a computing device of a Reviewing User; (Badawy, paragtraph 94, “To make the recommendation system more usable, the system can provide interpretations of the recommendations to users. For example, the peer ids, max similarity, output from ML model interpretation layer and other stats are generated as part of the output to help explain why a particular item was recommended. This information may also help a manager or other decision maker in the approval process.”, Badawy presents the access recommendation and its supporting information to a manager or other approval decision-maker. That manager or decision-maker is the claimed reviewing user, and the recommendation corresponds to the LLM-generated proposed permission set in the combination.)
(d2) receiving from said Reviewing User one or more clarification questions, (Badawy, paragraph 94, “To make the recommendation system more usable, the system can provide interpretations of the recommendations to users. For example, the peer ids, max similarity, output from ML model interpretation layer and other stats are generated as part of the output to help explain why a particular item was recommended. The user interface (UI) can feature a method to allow users to request such explanations to help with the approach and ML model’s transparency. As another example, if a user requests more information about a recommended entitlement, the system can provide detailed information about why access to a particular entitlement was recommended. Two examples of model interpretability tools which can be used for this purpose are LIME (local interpretable model-agnostic explanations) and SHAP (SHapley Additive exPlanations). Either interpretability tool can be used to interpret what features contributed to the generated recommendation, as one skilled in the art would understand. This information may help a user decide whether or not to request a recommended access item. This information may also help a manager or other decision maker in the approval process.”, A reviewing manager’s request for an explanation or additional information concerning the recommended entitlement is a clarification question concerning the proposed permission set.)
The rationale to combine Badawy with Ailwadi and Zaman is similar to as applied for claim 2 above.
Claims 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Ailawadi, in view of Zaman and Bogdanich et al., (US 2021/0051153 A1), hereafter referred to as Bogdanich.
Claim 9: Ailawadi and Zaman teaches the method of claim 1, Bogdanich, in the same field of machine learning, teaches the following which the above prior art fails to teach:
The computerized method of claim 1, wherein step (c) further comprises: automatically generating said LLM-based output that comprises definitions for automatic assignment of one or more pre-defined roles (Bogdanich, paragraphs 24 and 33, “The built-in roles file 130 includes a list of the roles available within the cloud service 102 and the actions allowed and not allowed for each role. The built-in role definition file is accessed (block 204). The built-in role definition file contains the roles within the system and the actions allowed and not allowed for each role.”, Bogdanich’s built-in role-definition file identifies roles already available in the cloud service and the permissions of each role. These built-in roles are predefined roles from which the system automatically selects and assigns one or more roles.)
that together include exactly said proposed minimal set of elevated permissions and no additional permissions. (Bogdanich, paragraphs 31 and 44, “The technique then maps the logged actions to a role having just the actions needed without any additional permissions. If one role has all the allowed actions of the service principal, then that role is selected as the recommended role (block 306). If no role exists that contains all the allowed actions, then the lowest-privilege role having the most allowed actions is selected and one or more additional low-privilege roles are selected to account for the remaining allowed actions (block 306). The recommended role may be a single role or a combination of roles.”, Bogdanich selects a single role or a combination of low-privilege roles that collectively provide the required actions and are expressly selected to avoid additional permissions. The selected predefined role combination therefore supplies the proposed minimal permission set.)
The Ailawadi-Zaman combination generates a task-specific least-privilege permission set and supports creation of a corresponding role. Bogdanich teaches implementing least privilege by selecting one or more built-in roles supported by a target cloud platform. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Bogdanich’s predefined-role selection when the target platform provides standardized roles because doing so would maintain platform compatibility, reduce unnecessary custom-role proliferation, and predictably provide the required actions with minimum excess privilege.
Claim 10: Ailawadi and Zaman teaches the method of claim 1, Ailwadi further teaches:
The computerized method of claim 1, wherein step (b) further comprises: feeding to said fine-tune LLM a signal indicating whether Custom Roles are supported or only Pre-Defined Roles are supported; (Ailwadi, paragraph 84, “Embodiments of the invention leverage large language models to instruct a software program or system which is meant to be invoked by a particular syntax, e.g. software is invoked by a function or method, e.g. an SQL database is invoked by an SQL query. In an input to the LLM the user provides a list of functions/sql queries/syntax along with the text in the form of prompts to support the LLM model to adapt to the proprietary information and syntax. The LLM model, which is fine tuned with the information provided, as explained below, then generates an output in the form of components of the function/query/syntax, e.g. name of function, attributes in the function and any other values which are relevant to invoke the functionality that function is responsible for in the intended software.”, Ailawadi feeds the LLM proprietary information identifying the functions, queries, and syntax supported by the target software. Whether the target software supports creation of custom roles or instead supports only assignment of predefined roles is target-software capability information of the same disclosed type. Supplying that information in the LLM prompt teaches the claimed support signal.)
Zaman further teaches:
if Custom Roles are supported, then the method comprises: automatically generating said LLM-based output that comprises definitions for automatic creation of a Custom Role that includes exactly said proposed minimal set of elevated permissions and no additional permissions; (Zaman, paragraph 26, “As shown by reference number 110, the role generation device may determine or generate one or more roles (e.g., one or more IAM roles) for the task based on the deployment pattern. For example, the one or more roles may be least privilege roles. For example, based on the information associated with the deployment pattern, the role generation device may determine what resources and/or services (e.g., associated with the cloud computing engine) are needed to perform the task. The role generation device may generate a role (e.g., an IAM role) for the task that only allows access to the resources and/or services that are needed to perform the task.”, Where custom role creation is available, Zaman generates a new task-specific IAM role limited to the permissions needed for the task. That generated least-privilege IAM role is the claimed custom role.)
Bogdanich, in the same field of machine learning, teaches the following which the above prior art fails to teach:
if Custom Roles are not supported and only Pre-Defined Roles are supported, then the method comprises: automatically generating said LLM-based output that comprises definitions for automatic assignment of one or more pre-defined roles that together include exactly said proposed minimal set of elevated permissions and no additional permissions. (Bogdanich, paragraphs 44, “When the roles are sorted by privilege, the number of logged actions that are made by the service principal in each role is determined (block 304). Then, the ordered role list is scanned from the lowest privilege role for the role that has the most allowed actions (block 306). If one role has all the allowed actions of the service principal, then that role is selected as the recommended role (block 306). If no role exists that contains all the allowed actions, then the lowest-privilege role having the most allowed actions is selected and one or more additional low-privilege roles are selected to account for the remaining allowed actions (block 306). The recommended role may be a single role or a combination of roles.”, Bogdanich selects from the roles supported by the platform and assigns either one predefined role or a combination of low-privilege predefined roles sufficient to provide all required actions. Selecting between Zaman’s custom-role implementation and Bogdanich’s predefined-role implementation according to the platform capabilities supplied to the LLM would have been a predictable implementation of the two known alternatives.)
Claims 11-14 are rejected under 35 U.S.C. 103 as being unpatentable over Ailawadi, in view of Zaman and Qin et al., (US 2024/0346256 A1), hereafter referred to as Qin.
Claim 11: Ailawadi and Zaman teaches the method of claim 1, Qin, in the same field of machine learning, teaches the following which Ailawadi and Zaman fails to teach:
The computerized method of claim 1, wherein step (b), of automatically feeding the REP as input into the fine-tuned LLM, comprises: (b1) enriching said REP by performing Retrieval Augmented Generation (RAG), and generating a RAG-enriched prompt; (Qin, paragraph 46, “Prompt generator 212 may generate a prompt for LLM 214 based on one or more of query 216, first feature vector 226, one or more of second feature vectors 228, indications 230, and/or augmentation information 232. For example, prompt generator 212 may generate an augmented prompt 236 that includes the original query, contextual information (e.g., the current webpage, the product or service of the current webpage, temporal information, location information, etc.), content information (e.g., the retrieved augmentation information 232), and a request to answer the original query based on the provided contextual information using the included content information. For example, prompt generator 212 may employ natural language processing (NLP) techniques to generate an augmented prompt 236 that requests LLM 214 to respond to query 216 based on contextual information using augmentation information 232. In embodiments, augmented prompt 236 may include, identify and/or link to augmentation information 232. Prompt generator 212 provides augmented prompt 236 to LLM 214.”, Qin retrieves augmentation information and combines it with the original query and contextual information to create an augmented prompt. Applying Qin’s process to the REP enriches the REP by RAG and generates the claimed RAG-enriched prompt.)
(b2) feeding said RAG-enriched prompt as input to the fine-tuned LLM. (Qin, paragraphs 46-47, “Prompt generator 212 provides augmented prompt 236 to LLM 214. In embodiments, LLM 214 receives augmented prompt 236 from prompt generator 212 and generates response 238. For example, LLM 214 may process augmented prompt 236 to generate a response 238 based on contextual information 215 using augmentation information 232.”, Qin expressly provides the augmented prompt to the LLM and causes the LLM to process that prompt to generate a response.)
Qin teaches using RAG to provide an LLM with domain-specific, entity-specific, product-specific, and updated information retrieved according to semantic similarity. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply Qin’s RAG mechanism to the Ailawadi-Zaman access-request system so that the permission recommendation is grounded in current service documentation and organizational authorization information that may not be represented in the LLM’s training data. The modification would predictably improve accuracy and reduce stale or unsupported permission recommendations.
Claim 12: Ailawadi, Zaman, and Qin teaches the method of claim 11, Zaman further teaches:
documentation items corresponding to a plurality of computerized services and cloud-based services. (Zaman, paragraphs 20-21, “The task may be associated with deploying (e.g., spinning up) infrastructure associated with the cloud computing environment. Deploying infrastructure may refer to the process of creating and configuring resources, such as virtual machines, storage, and networking, to support the deployment of an application or service. A permission may indicate one or more resources and/or services, associated with a cloud computing environment (e.g., associated with the cloud computing engine), that are permitted to be accessed when performing the task. The one or more resources and/or services may include an instance, a virtual machine, a container, a function (e.g., a lambda function), a bucket, and/or another resource or service associated with a cloud computing environment.”, Zaman identifies multiple computer and cloud services whose resources and permissions must be understood to construct the task role. It would have been obvious to use documentation for those disclosed cloud resources and services as Qin’s augmentation items so that the RAG system can accurately determine the permissions supported by the services.)
Qin, in the same field of machine learning further teaches:
The computerized method of claim 11, wherein enriching said REP is performed using a RAG unit that has access to a vector database that was constructed with embeddings from a plurality of documentation items (Qin, paragraphs 41-42, “Encoder 204 may process augmentation information text string 222 to generate a second feature vector 224 that represents the meaning of augmentation information text string 222. In embodiments, a second feature vector 224 may be generated for each piece of augmentation information in dataset(s) 112 and stored as second feature vectors 206 for future use. Encoder 204 may then generate embeddings for each token. In embodiments, the generated embeddings are numerical vectors that represent the meaning and context of the tokens. Encoder 204 may then aggregate the embeddings for each token to form a sentence embedding vector that represents the meaning of the text.”, Qin generates and stores an embedding vector for each piece of augmentation information in the dataset. The stored vectors and their corresponding source items form the claimed vector database constructed with embeddings from multiple documentation items.)
The rationale for combining Ailawadi and Zaman with Qin is similar to that as applied for claim 11 above.
Claim 13: Ailawadi, Zaman and Qin teaches the method of claim 12, Qin, in the same field of machine learning further teaches:
The computerized method of claim 12, wherein enriching said REP comprises: adding to said REP a context that includes selected segments of said documentation items that said RAG units determines to have semantic similarity to said REP beyond a pre-defined threshold level of semantic similarity. (Qin, paragraph 66, “In step 704, the file is pre-processed to generate processed augmentation information. For instance, pre processor 202 may process augmentation information 218 to generate an augmentation information text string 222 based on augmentation information 218. In embodiments, pre-processor 202 may generate a plurality of augmentation text strings 222 from each file containing augmentation information. For example, pre-processor 202 may process the file into a plurality of augmentation information text strings 222 based on a predetermined length, or based on segmentation information present in the file (e.g., by section, sub-section, and/or paragraph).”, Qin divides each source file into sections, subsections, paragraphs, or other text segments. Qin then retrieves selected augmentation information and includes it with the original query in the augmented prompt. Those retrieved source portions are the claimed selected documentation segments added as context to the REP.
Pargraph 61, “Flowchart 600A starts at step 602. In step 602, second feature vectors having a cosine similarity to the first feature vector that satisfies a predetermined condition with a predetermined threshold are determined. For instance, comparator 208 may calculate the cosine similarities between first feature vector 226 and a plurality of second feature vectors 228. Comparator 208 and/or retriever 210 may determine second feature vectors having calculated cosine similarities that satisfy a predetermined condition with a predetermined threshold. For example, comparator 208 and/or retriever 210 may determine second feature vectors having calculated cosine similarity values greater than a predetermined threshold (e.g., 0.8).”, Qin uses cosine similarity between the query vector and stored content vectors and selects vectors having similarity values greater than a predefined threshold. Because the vectors represent the meaning of the REP and documentation segments, the comparison is a semantic-similarity threshold comparison as claimed.)
The rationale for combining Ailawadi and Zaman with Qin is similar to that as applied for claim 11 above.
Claim 14: Ailawadi, Zaman and Qin teaches the method of claim 12, Ailawadi further teaches:
data-sources describing roles and permissions in said organization. (Ailawadi, paragraph 79, “Generate Embeddings. Use embedding techniques, e.g. Word Embeddings or Neural Embeddings, to represent access levels/job roles in a high-dimensional space. This creates vector representations that capture the relationships between different access levels or job roles respectively. Embeddings capture intricate relationships and nuances that may not be adequately represented in lower-dimensional spaces and are useful in finding insights such as semantic similarity which provide more precision while executing linear optimization and performing similarity analysis.”, Ailawadi’s source information expressly describes organizational access levels and job roles and embeds the relationships among them. Such access-level and role records are data sources describing roles and permissions.)
Qin, in the same field of machine learning further teaches:
The computerized method of claim 12, wherein enriching said REP is performed using a RAG unit that has access to a vector database that was constructed with embeddings from data-sources (Qin, paragraph 40, ““Pre-processor 202 may also receive augmentation information 218 from dataset(s) 112. As discussed above, augmentation information 218 may include, but is not limited to, domain-specific information, entity-specific information, product-specific information, recent information unavailable at generation of the large language model, and/or information changed after generation of the large language model. In embodiments, augmentation information 218 may further include metadata (e.g., product identifier) associated with augmentation information 218.”; paragraphs 41–42, “Encoder 204 may process augmentation information text string 222 to generate a second feature vector 224 that represents the meaning of augmentation information text string 222. In embodiments, a second feature vector 224 may be generated for each piece of augmentation information in dataset(s) 112 and stored as second feature vectors 206 for future use. Encoder 204 may then generate embeddings for each token. In embodiments, the generated embeddings are numerical vectors that represent the meaning and context of the tokens. Encoder 204 may then aggregate the embeddings for each token to form a sentence embedding vector that represents the meaning of the text.” Qin’s augmentation information 218 contained in dataset(s) 112 is interpreted as the claimed data-sources. Qin generates, for each piece of augmentation information in those data-sources, a second feature vector that represents the meaning of the information and stores the generated vectors as second feature vectors 206 for future retrieval. Qin further expressly identifies those numerical vectors as embeddings representing the meaning and context of the source text. Accordingly, the collection of stored second feature vectors 206, together with the corresponding augmentation information from dataset(s) 112, is interpreted as the claimed vector database constructed with embeddings from data-sources.)
The rationale for combining Ailawadi and Zaman with Qin is similar to that as applied for claim 11 above.
Claim 17 is rejected under 35 U.S.C. 103 as being unpatentable over Ailawadi, in view of Zaman and Parimi et al., (US 2017/0295181 A1), hereafter referred to as Parimi.
Claim 17: Ailawadi and Zaman teaches the method of claim 1, Ailawadi further teaches:
and automatically sending a message to the Requesting User indicating which permissions were elevated; (Ailwadi, paragraph 92, “On executing the function the final output is shown on the UI in a seamless way making sure that the user has to submit only natural language inputs and receive functionality or information in return based upon authorization of the user’s role.”, Ailawadi returns the result of the executed access function to the requesting user through the user interface. Because the system already possesses the generated permission or role information, including the permissions granted in the returned execution result would have been an obvious and predictable way to confirm the scope of the approved access.)
(d4) if in step (d2) a rejection signal was received from the Reviewing User, then: automatically sending a message to the Requesting User indicating that the REP was rejected. (Ailawadi, paragraph 82, “Once the deficiencies are identified and removed in the syntax of the function generated by the LLM, the feasibility of execution is decided by role based data and the access control block 55. This block makes sure that any resources involved in the execution of the function are legitimately entitled to be accessed by the user who is accessing the entire solution through the user interface (UI). In case the user is not authorized a message stating the same is shown at the UI, else the function syntax is sent to the next stage, i.e. function execution and server side functionality block 52 (short form execution block).”, Ailawadi automatically displays a message to the requesting user when the requested access-controlled operation is not authorized. Applying the same disclosed user-interface notification to the reviewer’s rejection branch informs the requesting user that the REP was rejected.)
Parimi, in the same field of machine learning, teaches the following which the above prior art fails to teach:
The computerized method of claim 1, further comprising: (d1) automatically providing the output of said fine-tuned LLM, in response to said REP, to a computing device of a Reviewing User; (d2) receiving from said Reviewing User either an approval or a rejection of the proposal of minimal set of elevated permissions that was generated by said fine-tuned LLM; (Parimi, paragraph 281, “The security management workflow 1804 includes receiving a low risk privilege 1104 request from user 104 and applying privilege management policies 1716 (e.g., security rule definition 1504 and/or security profile 1512) to auto-approve/grant the privilege and notifying the system user 904. The workflow also includes receiving a high risk privilege 1105 request from user 104, applying privilege management policies 1716, computing the risk score 210, notifying the system user 904, waiting for the system user’s 904 decision, and granting or denying the privilege depending on the system user’s response.”, Parimi provides the privilege request and decision information to a reviewing system user, waits for that user’s decision, and grants or denies the privilege according to the response. In the combination, the presented request includes the LLM-generated proposed minimal permission set, and the system user’s response is the claimed approval or rejection.)
(d3) if in step (d2) an approval signal was received from the Reviewing User, then: automatically invoking the minimal set of elevated permissions that were indicated in the proposal generated by said fine-tuned LLM, (Parimi, paragraph 281-282, “The user specific role is updated with the new privileges 1808 and the authorization server 1704 is notified in 1812. If it is a “no” in 1802 (i.e. the user specific role does not exist), a new user specific role is created with the new privileges in 1806 and the authorization server 1704 is notified in 1812. The new user specific role is assigned to the user in 1810 and the identity and authentication server 1702 is notified in 1816.”, After approval, Parimi updates an existing user-specific role or creates a new role containing the approved privileges and assigns the role to the user. That assignment automatically invokes the approved elevated permissions.)
Parimi teaches an established human-in-the-loop workflow for elevated or high-risk privilege requests, including notifying a reviewing system user, receiving the reviewer’s decision, and granting or denying the requested privilege. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to place Parimi’s approval workflow after the Ailawadi-Zaman permission recommendation because elevated access commonly requires human authorization before provisioning. The modification would predictably prevent automated recommendations from changing security-sensitive privileges without review while using Parimi’s approval mechanism for its ordinary purpose.
Claim 19 is rejected under 35 U.S.C. 103 as being unpatentable over Ailawadi, in view of Zaman and Brown et al., (Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., ... & Amodei, D. (2020). Language models are few-shot learners. Advances in neural information processing systems, 33, 1877-1901.), hereafter referred to as Brown.
Claim 19: Ailawadi and Zaman teaches the method of claim 1, Ailawadi further teaches:
The computerized method of claim 1, wherein the REP from the Requesting User is in a first natural language, and is processed by the fine-tuned LLM, (Ailwadi, paragraph 83-84, ““The end user can access the user interface 50 and provide inputs in the form of natural language text, voice, or image, e.g. via a search bar, submit button, a display areas, etc. The preprocessing block interacts with a module 57 that passes inputs and prompts to a large language model 56. The LLM model, which is fine tuned with the information provided, as explained below, then generates an output in the form of components of the function/query/syntax.”, Ailawadi receives a natural-language request and passes it to the fine-tuned LLM for processing, thereby teaching processing the REP in the requesting user’s first natural language.)
Brown, in the same field of machine learning, teaches the following which the above prior art fails to teach:
and wherein output from said fine-tuned LLM is provided to the Reviewing User in a second, different, natural language. (Brown, abstract, “For all tasks, GPT-3 is applied without any gradient updates or fine-tuning, with tasks and few-shot demonstrations specified purely via text interaction with the model. GPT-3 achieves strong performance on many NLP datasets, including translation, question-answering, and cloze tasks.”; page 6, section 3.3, “In collecting training data for GPT-3, we used the unfiltered distribution of languages reflected in internet text datasets (primarily Common Crawl)... By contrast, GPT-3 learns from a blend of training data that mixes many languages together.”, Brown establishes that a large language model processes text in multiple languages and performs translation. Applying that known LLM capability to Ailawadi’s interface permits the request to be received in one language and the resulting recommendation to be rendered for the reviewing user in a second language.)
Ailawadi teaches a natural-language LLM interface used by enterprise employees and decision-makers, while Brown teaches that a large language model performs translation through text interaction. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to use Brown’s known translation capability in Ailawadi’s interface so that requesters and reviewers using different languages could participate in the same privilege-review workflow. The modification would have applied a known LLM function for its ordinary purpose and would have predictably improved usability in multilingual organizations.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Subramaniam, P., & Krishnan, S. (2024). DePLOI: Applying NL2SQL to Synthesize and Audit Database Access Control. arXiv preprint arXiv:2402.07332.
US 12,177,223 B1
US 2024/0022607 A1
US 2021/0392142 A1
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HYUNGJUN B YI whose telephone number is (703)756-4799. The examiner can normally be reached M-F 9-5.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Usmaan Saeed can be reached on (571) 272-4046. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/H.B.Y./Examiner, Art Unit 2124
/DANIEL T PELLETT/Primary Examiner, Art Unit 2121
Prosecution Insights