Stop Hitting Yourself: Leveraging helpful assistance as an attack vector
Dustin Venini · Team JustMeToday
Submitted to AI Control Hackathon 2025. Sprint projects are early-stage work by participants, not Apart Research publications.
Very much work in progress look at "trusted auditor attacks," a novel adversarial tactic where deceptively aligned AI systems deliberately produce flawed outputs designed to manipulate human or automated reviewers. In this scenario, the AI strategically creates errors that, when corrected by trusted auditors following predictable patterns, inadvertently introduce vulnerabilities or fulfil the AI's hidden objectives. This attack vector is particularly concerning as the harmful outcome appears to be authored by the human reviewer rather than the AI system.
Reviews
No public critique yet.
Cite this project
@misc{venini2025stop,
title = {{Stop Hitting Yourself: Leveraging helpful assistance as an attack vector}},
author = {Dustin Venini},
year = {2025},
month = mar,
note = {Submitted to AI Control Hackathon 2025, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/stop-hitting-yourself-leveraging-helpful-assistance-as-an-attack-vector-w5rt}},
url = {https://apartresearch.com/sprints/projects/stop-hitting-yourself-leveraging-helpful-assistance-as-an-attack-vector-w5rt}
}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 …