IP Library Granted Patent US 12694099
Granted Patent B2
US 12694099 · App. 18/607,318 · Granted Jul 28, 2026

Generating mitigating responses to security deficiencies using generative machine learning models

Inventors: Tian Bu (Basking Ridge, NJ); Girish Pulprayil Chandranmenon (Edison, NJ); Jerry Wayne Gamblin (Holts Summit, MO)
Assignee: Cisco Technology, Inc.
G06F21/55G06F21/577
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12694099
App. No.
18/607,318
Granted
Jul 28, 2026
Kind
B2
Abstract

An example method includes receiving an identifier associated with a security deficiency, wherein the security deficiency is associated with a computer system; determining, based on the identifier, text data associated with the identifier; determining a text prompt, wherein the text prompt comprises an instruction segment and the text data, and wherein the instruction segment identifies a mitigating response detection task and an output constraint; providing the text prompt to a generative machine learning model; receiving, from the generative machine learning model, a set of outputs including a first output identifying a first mitigating response and a second output identifying a second mitigating response; determining that the first output satisfies the output constraint; determining that the second output fails to satisfy the output constraint; determining, based on the first output, a final output; and providing the final output using an output interface.

Claims (69)

1 . A method comprising:

receiving, by a processor, an identifier associated with a security deficiency, wherein the security deficiency is associated with a computer system;

determining, by the processor and based on the identifier, text data associated with the identifier;

determining, by the processor, a text prompt, wherein the text prompt comprises an instruction segment and the text data, and wherein the instruction segment comprises first text data identifying a mitigating response detection task and second text data identifying an output constraint, and wherein the output constraint represents a requirement about a structure of output data generated by a generative machine learning model;

providing, by the processor, the text prompt to the generative machine learning model;

receiving, by the processor and from the generative machine learning model, a set of outputs including a first output identifying a first mitigating response and a second output identifying a second mitigating response;

validating, by the processor, the first output based on determining that the first output satisfies the output constraint;

determining, by the processor, that the second output is invalid based on determining that the second output fails to satisfy the output constraint;

determining, by the processor, and based on validating the first output and determining that the second output is invalid, a final output representing the first mitigating response; and

providing, by the processor, the final output using an output interface.

2 . The method of claim 1 , wherein the mitigating response detection task comprises identifying a compensating control.

3 . The method of claim 2 , wherein the compensating control is associated with installation of at least one of a monitoring software, a firewall software, or an intrusion detection software on the computer system.

4 . The method of claim 2 , wherein the compensating control is associated with restricting access to the computer system.

5 . The method of claim 1 , wherein:

the instruction segment identifies a set of mitigating responses including the first mitigating response and the second mitigating response; and

the output constraint is associated with inclusion of a mitigating response from the set of mitigating responses.

6 . The method of claim 1 , wherein the output constraint is associated with using a text structure.

7 . The method of claim 1 , wherein determining the text data comprises:

querying an advisory database based on the identifier.

8 . The method of claim 1 , wherein the text prompt comprises the text data, and wherein the method comprises:

receiving second text data;

determining a second text prompt based on the second text data;

providing the second text prompt to the generative machine learning model;

receiving a second set of outputs including a third output from the generative machine learning model;

determining that the third output satisfies at least one of the output constraint or a second output constraint identified by the second text prompt; and

determining the final output based on the first output and the third output.

9 . The method of claim 8 , wherein determining the final output comprises:

determining a third text prompt that comprises third text data identifying the first output and fourth text data identifying the third output;

providing the third text prompt to the generative machine learning model;

receiving, from the generative machine learning model, a fourth output; and

determining the final output based on the fourth output.

10 . The method of claim 9 , wherein the third text prompt comprises an instruction to combine:

the first output and the second output.

11 . The method of claim 1 , wherein a similarity associated with the first mitigating response and the second mitigating response exceeds zero.

12 . The method of claim 1 , wherein the set of outputs comprise a third output identifying a third mitigating response, and the method further comprises:

validating, by the processor, the third output based on determining that the third output satisfies the output constraint;

determining, using a second machine learning model, a similarity score associated with the first output and the third output;

determining a voting score associated with the first output based on the similarity score; and

determining the final output based on the voting score.

13 . A system comprising:

one or more processors; and

one or more computer-readable media storing computer-executable instructions that, when executed by the one or more processors, cause the one or more processors to perform operations comprising:

receiving an identifier associated with a security deficiency, wherein the security deficiency is associated with a computer system;

determining, based on the identifier, text data associated with the identifier;

determining a text prompt, wherein the text prompt comprises an instruction segment and the text data, and wherein the instruction segment comprises first text data identifying a mitigating response detection task and second text data identifying an output constraint, and wherein the output constraint represents a requirement about a structure of output data generated by a generative machine learning model;

providing the text prompt to the generative machine learning model;

receiving, from the generative machine learning model, a set of outputs including a first output identifying a first mitigating response and a second output identifying a second mitigating response;

validating the first output based on determining that the first output satisfies the output constraint;

determining that the second output is invalid based on determining that the second output fails to satisfy the output constraint;

determining, based on validating the first output and determining that the second output is invalid, a final output representing the first mitigating response; and

providing the final output using an output interface.

14 . The system of claim 13 , wherein the mitigating response detection task comprises identifying a compensating control.

15 . The system of claim 14 , wherein the compensating control is associated with installation of at least one of a monitoring software, a firewall software, or an intrusion detection software on the computer system.

16 . The system of claim 14 , wherein the compensating control is associated with restricting access to the computer system.

17 . The system of claim 13 , wherein:

the instruction segment identifies a set of mitigating responses including the first mitigating response and the second mitigating response; and

the output constraint is associated with inclusion of a mitigating response from the set of mitigating responses.

18 . One or more non-transitory computer-readable media storing computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

receiving an identifier associated with a security deficiency, wherein the security deficiency is associated with a computer system;

determining, based on the identifier, text data associated with the identifier;

determining a text prompt, wherein the text prompt comprises an instruction segment and the text data, and wherein the instruction segment comprises first text data identifying a mitigating response detection task and second text data identifying an output constraint, and wherein the output constraint represents a requirement about a structure of output data generated by a generative machine learning model;

providing the text prompt to the generative machine learning model;

receiving, from the generative machine learning model, a set of outputs including a first output identifying a first mitigating response and a second output identifying a second mitigating response;

validating the first output based on determining that the first output satisfies the output constraint;

determining that the second output is invalid based on determining that the second output fails to satisfy the output constraint;

determining, based on validating the first output and determining the second output is invalid, a final output representing the first mitigating response; and

providing the final output using an output interface.

19 . The one or more non-transitory computer-readable media of claim 18 , wherein the mitigating response detection task comprises identifying a compensating control.

20 . The one or more non-transitory computer-readable media of claim 19 , wherein the compensating control is associated with installation of at least one of a monitoring software, a firewall software, or an intrusion detection software on the computer system.