Safety Metric and Prompt Engineering for Red Team
Zhu Liang · Team Paradite
Submitted to AI Control Hackathon 2025. Sprint projects are early-stage work by participants, not Apart Research publications.
In the ControlArena environment, we have a default attack policy that is very basic. We want to explore how to improve the attack policy using basic prompt engineering techniques.
We also realized that the security metric evaluation in the environment is not closely following the foundational paper (Greenblatt et al., 2023), so we made a change to more closely approximate the security metric in the paper.
Reviews
No public critique yet.
Cite this project
@misc{liang2025safety,
title = {{Safety Metric and Prompt Engineering for Red Team}},
author = {Zhu Liang},
year = {2025},
month = mar,
note = {Submitted to AI Control Hackathon 2025, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/safety-metric-and-prompt-engineering-for-red-team-q2d6}},
url = {https://apartresearch.com/sprints/projects/safety-metric-and-prompt-engineering-for-red-team-q2d6}
}More from AI Control Hackathon 2025
- 1st place by peer reviewView project: Can models use their Chain-of-Thought to attack overseers?
Can models use their Chain-of-Thought to attack overseers?
This project explores vulnerabilities in AI evaluation mechanisms, specifically focusing on how AI agents might influence their overseeing AI judges through manipulative instructions embedded within their …
- 2nd place by peer reviewView project: Schelling coordination via agentic loops
Schelling coordination via agentic loops
Focal
Capability to collude undermines untrusted monitoring protocols. We propose a novel way to evaluate acausal coordination capabilities of LLMs via repeated sampling in causally separated agentic loops. We provide an …
- 3rd place by peer reviewView project: Model Models: Simulating a Trusted Monitor
Model Models: Simulating a Trusted Monitor
Charlie's Angels
We offer initial investigations into whether the untrusted model can 'simulate' the trusted monitor: is U able to successfully guess what suspicion score T will assign in the APPS setting? We also offer a clean, modular …