# US 2025/0097237 A1 · Fully automated pen testing with security policy correction using generative llm

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2025/0097237 A1
- Document ID: US20250097237A1
- Status: Published application
- Kind: A1
- Application number: 18/470,021
- Filed: 2023-09-19
- Published: 2025-03-20
- Earliest claimed: 2023-09-19
- Assignees: Cisco Technology, Inc.
- Applicants: Cisco Technology, Inc.
- Inventors: Vincent Parla
- Web page: https://patentgrep.com/patent/US20250097237A1

## CPC

- H04L 63/1416: {Event detection, e.g. attack signature detection}
- H04L 41/16: using machine learning or artificial intelligence
- H04L 63/1433: {Vulnerability analysis}
- H04L 63/20: {for managing network security; network security policies in general (filtering policies H04L63/0227)}
- G06F 11/1469: {Backup restoration techniques}
- G06F 11/302: {where the computing system component is a software system}
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

In one aspect, a method for penetration testing for optimization of network security policies is disclosed. The method includes determining, by a security management service, that one or more cybersecurity threats successfully penetrated a security service protecting a pseudo-target in a penetration testing environment, analyzing, by the security management service, the one or more cybersecurity threats that successfully penetrated the security service to characterize the one or more cybersecurity threats, and generating, by the security management service, an update of a policy used by the security service that would prevent the one or more cybersecurity threats from penetrating the security service based on the analysis of the one or cybersecurity threats.

## Claims

### Claim 1

A method comprising:
determining, by a security management service, that one or more cybersecurity threats successfully penetrated a security service protecting a pseudo-target in a penetration testing environment;
analyzing, by the security management service, the one or more cybersecurity threats that successfully penetrated the security service to characterize the one or more cybersecurity threats; and
generating, by the security management service, an update of a policy used by the security service that would prevent the one or more cybersecurity threats from penetrating the security service based on the analysis of the one or more cybersecurity threats.

### Claim 2 (depends on 1)

The method of claim 1, wherein the analyzing the one or more cybersecurity threats that successfully penetrated the security service and the generating the update of the policy used by the security service is performed using a large language model (LLM).

### Claim 3 (depends on 2)

The method of claim 2, wherein the LLM inspects a current policy used by the security service, and the LLM is prompted to analyze the one or more cybersecurity threats, and to generate the update of the policy.

### Claim 4 (depends on 2)

The method of claim 2, further comprising:
generating additional variants related to the one or more cybersecurity threats that successfully penetrated the security service;
deploying at least a portion of the additional variants to observe the additional variants in the penetration testing environment;
analyzing results of the additional variants in the penetration testing environment; and
applying the analysis of the results of the additional variants in the penetration testing environment aid the generating the update of the policy used by the security service.

### Claim 5 (depends on 4)

The method of claim 4, further comprising:
deploying at least a portion of the additional variants to observe the additional variants in the penetration testing environment, wherein the penetration testing environment has been updated with the generated policy used by the security service that is intended to prevent the one or more cybersecurity threats from penetrating the security service.

### Claim 6 (depends on 2)

The method of claim 2, wherein the security management service is configured to instruct the LLM to:
identify a plurality of patterns related to successful previous penetrations of the one or more cybersecurity threats; and
provide a plurality of recommendations based on the plurality of patterns, wherein the plurality patterns identify patterns in one or more attack vectors.

### Claim 7 (depends on 1)

The method of claim 1, wherein the security management service employs classifications of the one or more cybersecurity threats into one or more threat categories based on characterizations of each of the one or more cybersecurity threats, and automatically generates a threat score by evaluating a ranking of these classifications of the one or more cybersecurity threats.

### Claim 8 (depends on 7)

The method of claim 7, wherein the threat score quantifies a risk level posed by each of the one or more cybersecurity threats and prioritizes one or more preventative actions to reduce the penetration of the one or more cybersecurity threats based on the threat score.

### Claim 9

A network device comprising:
one or more memories having computer-readable instructions stored therein; and
one or more processors configured to execute the computer-readable instructions to:
determine, by a security management service, that one or more cybersecurity threats successfully penetrated a security service protecting a pseudo-target in a penetration testing environment;
analyze, by the security management service, the one or more cybersecurity threats that successfully penetrated the security service to characterize the one or more cybersecurity threats; and
generate, by the security management service, an update of a policy used by the security service that would prevent the one or more cybersecurity threats from penetrating the security service based on the analysis of the one or cybersecurity threats.

### Claim 10 (depends on 9)

The network device of claim 9, wherein the analyzing the one or more cybersecurity threats that successfully penetrated the security service and the generating the update of the policy used by the security service is performed using a large language model (LLM).

### Claim 11 (depends on 10)

The network device of claim 10, wherein the LLM is aware of a current policy used by the security service, and the LLM is prompted to analyze the one or more cybersecurity threats, and to generate the update of the policy.

### Claim 12 (depends on 9)

The network device of claim 9, further comprising:
generating additional variants related to the one or more cybersecurity threats that successfully penetrated the security service;
deploying at least a portion of the additional variants to observe the additional variants in the penetration testing environment;
analyzing results of the additional variants in the penetration testing environment; and
applying the analysis of the results of the additional variants in the penetration testing environment aid the generating the update of the policy used by the security service.

### Claim 13 (depends on 12)

The network device of claim 12, further comprising:
deploying at least a portion of the additional variants to observe the additional variants in the penetration testing environment, wherein the penetration testing environment has been updated with the generated policy used by the security service that is intended to prevent the one or more cybersecurity threats from penetrating the security service.

### Claim 14 (depends on 9)

The network device of claim 9, wherein the security management service is configured to instruct an LLM to:
identify a plurality of patterns related to successful previous penetrations of the one or more cybersecurity threats; and
provide a plurality of recommendations based on the plurality of patterns, wherein the plurality patterns identify patterns in one or more attack vectors.

### Claim 15 (depends on 9)

The network device of claim 9, wherein the security management service employs classifications of the one or more cybersecurity threats into one or more threat categories based on characterizations of each of the one or more cybersecurity threats, and automatically generates a threat score by evaluating a ranking of these classifications of the one or more cybersecurity threats.

### Claim 16 (depends on 15)

The network device of claim 15, wherein the threat score quantifies a risk level posed by each of the one or more cybersecurity threats and prioritizes one or more preventative actions to reduce the penetration of the one or more cybersecurity threats based on the threat score.

### Claim 17

A non-transitory computer-readable storage medium comprising computer-readable instructions, which when executed by one or more processors of a network appliance, cause the network appliance to:
determine, by a security management service, that one or more cybersecurity threats successfully penetrated a security service protecting a pseudo-target in a penetration testing environment;
analyze, by the security management service, the one or more cybersecurity threats that successfully penetrated the security service to characterize the one or more cybersecurity threats; and
generate, by the security management service, an update of a policy used by the security service that would prevent the one or more cybersecurity threats from penetrating the security service based on the analysis of the one or cybersecurity threats.

### Claim 18 (depends on 17)

The non-transitory computer-readable storage medium of claim 17, wherein the analyzing the one or more cybersecurity threats that successfully penetrated the security service and the generating the update of the policy used by the security service is performed using a large language model (LLM).

### Claim 19 (depends on 18)

The non-transitory computer-readable storage medium of claim 18, wherein the LLM is aware of a current policy used by the security service, and the LLM is prompted to analyze the one or more cybersecurity threats, and to generate the update of the policy.

### Claim 20 (depends on 17)

The non-transitory computer-readable storage medium of claim 17, wherein the one or more processors are further configured to:
generate additional variants guided by the analyzing of the one or more cybersecurity threats that successfully penetrated the security service.
