AI Honeypot
Reworr
Submitted to Agent Security Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.
The project designed to monitor AI Hacking Agents in the real world using honeypots with prompt injections and temporal analysis.
Reviews
A creative approach to using honeypots with prompt injection techniques is both fascinating and effective. The project’s ability to gather real-world data on AI behavior is a game-changer, offering a unique glimpse into emerging cybersecurity challenges. It’s a true standout in the field of agent safety!
Great deliverables. The website containing the explainer and paper together with the animated slides improve the communication aspect significantly.• The idea temporal analysis to distinguish LLM agents from humans based on response times is particularly good.
I find your project to be an innovative and timely approach to tackling AI-driven cybersecurity threats. Your integration of real-time metrics through a public-facing dashboard is impressive, offering transparency and practical insights to the cybersecurity community. One area for improvement could be around increasing the dataset of AI-driven hacking attempts, which seems currently limited. This could help in refining detection methods further and provide more robust insights.
This is a great idea on securing systems from AI agents’s attacks. I would love to see two things in future releases: (a) False positive and False negative rates of Honeypot, (b): I believe LLM agents are becoming quite smart w.r.t. prompt injections. As such, do the prompt injection methods work over time?
Cite this project
@misc{reworr2024ai,
title = {{AI Honeypot}},
author = {Reworr},
year = {2024},
month = oct,
note = {Submitted to Agent Security Hackathon, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/ai-honeypot}},
url = {https://apartresearch.com/sprints/projects/ai-honeypot}
}More from Agent Security Hackathon
- 1st place by peer reviewView project: Diamonds are Not All You Need
Diamonds are Not All You Need
Diamonds are Not All You Need
This project tests an AI agent in a straightforward alignment problem. The agent is given creative freedom within a Minecraft world and is tasked with transforming a 100x100 radius of the world into diamond. It is …
- View project: Cross-model surveillance for emails handling
Cross-model surveillance for emails handling
Fluffy Vin
A system that implements cross-model security checks, where one AI agent (Agent A) interacts with another (Agent B) to ensure that potentially harmful actions are caught and mitigated before they can be executed. …
- View project: Inference-Time Agent Security
Inference-Time Agent Security
Inference-Time Agent Security
We take a first step towards automating model building for symbolic checking (eg formal verification, PDDL) of LLM systems.