Cross-model surveillance for emails handling
Le Ngoc Mai · Team Fluffy Vin
Submitted to Agent Security Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.
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. Specifically, Agent A is responsible for generating and sending emails, while Agent B reads these emails to determine whether they contain spam or malicious content. If Agent B detects that an email is spam, it triggers a shutdown of Agent A, effectively preventing any further potentially harmful actions.
Reviews
This project reminds me of Generative Adversarial Networks (GANs) or traditional spam detectors. Given the increasing prevalence of LLM-generated content, there's definitely a need for more advanced spam detection systems.I commend your thoroughness in actually implementing agent-to-agent email communication, though this level of detail wasn't strictly necessary for the hackathon. You could have focused more narrowly on spam detection. Nonetheless, you've identified a solid problem statement and developed a good project overall.
Proactive AI safety in action! The concept of using one agent to oversee another is clever and effective, creating a strong safety net against potential misuse. It’s a thoughtful and well-executed solution that demonstrates the importance of layered security in AI systems. A great step forward for maintaining safe and secure AI interactions!
pretty basic. does not address the hard part of the problem (how to get an agent interaction infrastructure that can be enforced).
Detecting spam emails is a solid use case showcasing the importance of the idea and the effectiveness of the approach. However, it would be interesting to expand the methodology to challenging cases that are arising in the AI agents landscape. Looking forward to how you expand the scope of this work.
Cite this project
@misc{mai2024crossmodel,
title = {{Cross-model surveillance for emails handling}},
author = {Le Ngoc Mai},
year = {2024},
month = oct,
note = {Submitted to Agent Security Hackathon, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/cross-model-surveillance-for-emails-handling}},
url = {https://apartresearch.com/sprints/projects/cross-model-surveillance-for-emails-handling}
}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: 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.
- View project: Cop N' Shop
Cop N' Shop
Marketplace Watchdogs
This paper proposes the development of AI Police Agents (AIPAs) to monitor and regulate interactions in future digital marketplaces, addressing challenges posed by the rapid growth of AI-driven exchanges. Traditional …