IP Library Granted Patent US 12,423,441
Granted Patent B2
US 12,423,441 · App. 18/393,487 · Granted Sep 23, 2025

Method for using generative large language models (LLM) for cybersecurity deception and honeypots

Inventors: David Arthur McGrew (Poolesville, MD); Hugo Mike Latapie (Long Beach, CA); Blake Anderson (Chapel Hill, NC)
Assignee: Cisco Technology, Inc.
H04L63/1433G06F11/3476G06F16/334G06F16/345G06F16/9024G06F21/31G06F21/552G06F21/563G06F21/577H04L63/1425H04L63/145H04L63/1483H04L63/1491
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 12,423,441
App. No.
18/393,487
Granted
Sep 23, 2025
Kind
B2
Abstract

In one aspect, a method for enhancing cybersecurity using Large Language Model (LLM)-generated honeypot schemes, the method includes generating a plurality of deceptive information using an LLM, configured to attract and engage potential attackers, where the plurality of deceptive information includes one or more characteristics referencing vulnerabilities of a network, continuously monitoring for interactions initiated by an interacting party with one or more components of the generated deceptive information, where the interaction is identified as a potential threat to the network, in response to detection of an interaction identified as a potential threat, extracting interaction data associated with the interacting party retrieved during the interaction, and retraining the LLM with the interaction data to create more effective honeypots.

Claims (49)

1. A method for enhancing cybersecurity using Large Language Model (LLM)-generated honeypot schemes, the method comprising:

generating a plurality of deceptive information using an LLM, configured to attract and engage potential attackers, wherein the plurality of deceptive information comprises one or more characteristics referencing vulnerabilities of a network;

continuously monitoring for a first interaction initiated by an interacting party with one or more components of the generated deceptive information, wherein the first interaction is identified as a potential threat to the network;

in response to detection of an interaction identified as the potential threat, extracting interaction data associated with the interacting party retrieved during the first interaction; and

retraining the LLM with the interaction data to create more effective honeypot schemes.

2. The method of claim 1 , wherein the LLM-generated honeypot schemes further include a honey pot service configured to prompt the LLM to generate the plurality of deceptive information in accordance with a predetermined script, ensuring consistency with the plurality deceptive information.

3. The method of claim 1 , wherein the plurality of deceptive information includes one or more fabricated user accounts, files, and administrator accounts configured to engage the potential attackers.

4. The method of claim 1 , further comprising:

tokenization of generated deceptive information to provide realistic HTTP responses in response to interactions initiated by the interacting party.

5. The method of claim 1 , wherein, in response to the detection of the interaction identified as the potential threat, the method further includes:

generating one or more remedial measures and policies for the network based on the interaction data, enhancing network security.

6. The method of claim 1 , wherein the continuously monitoring includes generating one or more predictions of a type of interaction to engage the potential threat associated with the one or more characteristics referencing vulnerabilities of the network.

7. The method of claim 1 , wherein the continuously monitoring includes:

generating one or more contextual labels in accordance with contextual data related to the interactions to distinguish and identify threatening interactions from non-threatening interactions; and

retraining the LLM with the one or more contextual labels improving an accuracy level of potential threat detection and effectiveness of the honeypot schemes.

8. The method of claim 1 , further comprising:

training the LLM by leveraging malware data stored in a storage, wherein the LLM references the malware data to identify potential threats in the network.

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:

generate a plurality of deceptive information using an LLM, configured to attract and engage potential attackers, wherein the plurality of deceptive information comprises one or more characteristics referencing vulnerabilities of a network;

continuously monitor for a first interaction initiated by an interacting party with one or more components of the generated deceptive information, wherein the first interaction is identified as a potential threat to the network;

in response to detection of an interaction identified as the potential threat, extract interaction data associated with the interacting party retrieved during the first interaction; and

retrain the LLM with the interaction data to create more effective honeypot schemes.

10. The network device of claim 9 , wherein the instructions further cause the one or more processors to:

tokenize generated deceptive information to provide realistic HTTP responses in response to interactions initiated by the interacting party.

11. The network device of claim 9 , wherein in response to the detection of an interaction identified as the potential threat, the instructions further cause the processor to:

generating one or more remedial measures and policies for the network based on the interaction data, enhancing network security.

12. The network device of claim 9 , wherein the continuously monitoring includes:

generating one or more predictions of a type of interaction to engage the potential threat associated with the one or more characteristics referencing vulnerabilities of the network.

13. The network device of claim 9 , wherein the continuously monitoring includes:

generating one or more contextual labels in accordance with contextual data related to the interactions to distinguish and identify threatening interactions from non-threatening interactions; and

retraining the LLM with the one or more contextual labels improving an accuracy level of potential threat detection and effectiveness of the honeypot schemes.

14. 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:

generate a plurality of deceptive information using an LLM, configured to attract and engage potential attackers, wherein the plurality of deceptive information comprises one or more characteristics referencing vulnerabilities of a network;

continuously monitor for a first interaction initiated by an interacting party with one or more components of the generated deceptive information, wherein the first interaction is identified as a potential threat to the network;

in response to detection of an interaction identified as the potential threat, extract interaction data associated with the interacting party retrieved during the interaction; and

retrain the LLM with the interaction data to create more effective honeypot schemes.

15. The non-transitory computer-readable storage medium of claim 14 , wherein the LLM-generated honeypot schemes further include a honey pot service configured to prompt the LLM to generate the plurality of deceptive information in accordance with a predetermined script, ensuring consistency with the plurality deceptive information.

16. The non-transitory computer-readable storage medium of claim 14 , wherein the one or more processors are further configured to:

tokenize generated deceptive information to provide realistic HTTP responses in response to interactions initiated by the interacting party.

17. The non-transitory computer-readable storage medium of claim 14 , wherein the one or more processors are further configured to:

generate one or more remedial measures and policies for the network based on the interaction data, enhancing network security.

18. The non-transitory computer-readable storage medium of claim 14 , wherein the continuously monitor includes generate one or more predictions of a type of interaction to engage the potential threat associated with the one or more characteristics referencing vulnerabilities of the network.

19. The non-transitory computer-readable storage medium of claim 14 , wherein the one or more processors are further configured to:

generate one or more contextual labels in accordance with contextual data related to the interactions to distinguish and identify threatening interactions from non-threatening interactions; and

retrain the LLM with the one or more contextual labels improving an accuracy level of potential threat detection and effectiveness of the honeypot schemes.

20. The non-transitory computer-readable storage medium of claim 14 , wherein the one or more processors are further configured to:

train the LLM by leveraging malware data stored in a storage, wherein the LLM references the malware data to identify potential threats in the network.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Dec 21, 2023
From: MCGREW, DAVID ARTHUR; LATAPIE, HUGO MIKE; ANDERSON, BLAKE
To: CISCO TECHNOLOGY, INC.
Reel/Frame 065936/0950 →
Continuity (2)
Provisional Application 63493552 · Mar 31, 2023
Related Publication 20240333765A1 · Oct 3, 2024
References Cited (10)
US 10320841B1 · Allen · 2019 [cited by examiner]
US 10379995B1 · Walters · 2019 [cited by examiner]
US 11050787B1 · Sharifi Mehr · 2021 [cited by examiner]
US 11057429B1 · Sellers · 2021 [cited by applicant]
US 20180262529A1 · Allen · 2018 [cited by examiner]
US 20210067553A1 · Ries et al. · 2021 [cited by applicant]
US 20220053022A1 · Wu et al. · 2022 [cited by applicant]
US 20220103582A1 · Kidney · 2022 [cited by examiner]
CN 114363093 · 2022 [cited by applicant]
CN 115987610 · 2023 [cited by applicant]