# Attack Selection in Agentic AI Control Evaluations Meaningfully Decreases Safety

arXiv: https://arxiv.org/abs/2606.06529 · HTML: https://mlsanity.com/paper/2606.06529/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use.
- Builds on 12 of its 20 references that are in the map.
- First posted on arXiv 3 Jun 2026.

## Placed in

- LLM agents and tool use / [Agent threats, guardrails and governance](https://mlsanity.com/agents/agent-threats-guardrails-and-governance/)

## Builds on (most cited first, 12 of 12)

- [AI Control: Improving Safety Despite Intentional Subversion](https://mlsanity.com/paper/2312.06942/) (2023)
- [SHADE-Arena: Evaluating Sabotage and Monitoring in LLM Agents](https://mlsanity.com/paper/2506.15740/) (2025)
- [Sabotage Evaluations for Frontier Models](https://mlsanity.com/paper/2410.21514/) (2024)
- [Ctrl-Z: Controlling AI Agents via Resampling](https://mlsanity.com/paper/2504.10374/) (2025)
- [A sketch of an AI control safety case](https://mlsanity.com/paper/2501.17315/) (2025)
- [Games for AI Control: Models of Safety Evaluations of AI Deployment Protocols](https://mlsanity.com/paper/2409.07985/) (2024)
- [Adaptive Deployment of Untrusted LLMs Reduces Distributed Threats](https://mlsanity.com/paper/2411.17693/) (2024)
- [Adaptive Attacks on Trusted Monitors Subvert AI Control Protocols](https://mlsanity.com/paper/2510.09462/) (2025)
- [BashArena: A Control Setting for Highly Privileged AI Agents](https://mlsanity.com/paper/2512.15688/) (2025)
- [Subversion Strategy Eval: Can language models statelessly strategize to subvert control protocols?](https://mlsanity.com/paper/2412.12480/) (2024)
- [Optimizing AI Agent Attacks With Synthetic Data](https://mlsanity.com/paper/2511.02823/) (2025)
- [Attack Selection Reduces Safety in Concentrated AI Control Settings against Trusted Monitoring](https://mlsanity.com/paper/2602.04930/) (2026)

## Built on by (most cited first, 1 of 1)

- [Reliability Theory for AI Control](https://mlsanity.com/paper/2609.26419/) (2026)

## Abstract (arXiv)

An attacker that strategically chooses when to attack is much harder to catch than one that attacks indiscriminately. AI control is a safety framework for deploying capable but untrusted AI agents under the oversight of a weaker, trusted monitor and a limited human audit budget. Control evaluations stress-test these protocols by pitting a red-team attack policy against the blue-team monitor, but current evaluations typically assume attackers that do not strategically select when to attack. We study this capability, attack selection, in agentic settings by decomposing attack decisions into a start policy, which decides when an attacker should attack, and a stop policy, which decides when an attacker should abort an ongoing attack. Across two agentic settings, BashArena and LinuxArena, both policies substantially lower measured empirical safety without changing the underlying attack capability. At a 1% audit budget, our start policy reduces safety by 20pp on both BashArena and LinuxArena, and our stop policy reduces safety by 20pp on BashArena and 28pp on LinuxArena. These reductions should be interpreted as upper bounds on the effect of attack selection. Existing control evaluations may therefore yield overly optimistic safety estimates against selective attackers. We recommend that future evaluations, system cards, and safety cases elicit attack selection to produce more realistic safety estimates.