# AI Security Policy Should Assess Systems, Not Only Models

arXiv: https://arxiv.org/abs/2605.09504 · HTML: https://mlsanity.com/paper/2605.09504/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use.
- Builds on 8 of its 30 references that are in the map.
- First posted on arXiv 10 May 2026.

## Placed in

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

## Builds on (most cited first, 8 of 8)

- [HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal](https://mlsanity.com/paper/2402.04249/) (2024)
- [Tree of Attacks: Jailbreaking Black-Box LLMs Automatically](https://mlsanity.com/paper/2312.02119/) (2023)
- [AI Sandbagging: Language Models can Strategically Underperform on Evaluations](https://mlsanity.com/paper/2406.07358/) (2024)
- [Vulnerability Detection with Code Language Models: How Far Are We?](https://mlsanity.com/paper/2403.18624/) (2024)
- [The Jailbreak Tax: How Useful are Your Jailbreak Outputs?](https://mlsanity.com/paper/2504.10694/) (2025)
- [The Elicitation Game: Evaluating Capability Elicitation Techniques](https://mlsanity.com/paper/2502.02180/) (2025)
- [Synthesizing Multi-Agent Harnesses for Vulnerability Discovery](https://mlsanity.com/paper/2604.20801/) (2026)
- [Confusion is the Final Barrier: Rethinking Jailbreak Evaluation and Investigating the Real Misuse Threat of LLMs](https://mlsanity.com/paper/2508.16347/) (2025)

## Abstract (arXiv)

We show that the instruments used to measure AI offensive capability fail in two ways: (i) they overstate harm, and (ii) they credit the model with capability that belongs to the surrounding system. We argue that restricting access to a model is therefore necessary but not sufficient and that policy and procurement also need system-level, harm-grounded capability assessment. In June 2026, two frontier models were suspended under US export controls, reportedly prompted by a jailbreak that asked a model to read a codebase and fix its flaws. This finding measured an elicitation system of model, prompt, and task. We support our argument with a study of an open-source framework in which lightweight large language model (LLM) agents coordinate through shared memory and evolutionary optimization, providing two pieces of evidence. First, jailbreak metrics overstate harm: over 225 swarm-generated attacks per target, LLM-as-judge scoring rated Claude Sonnet 4 compromised in 40% of attacks, yet manual verification found actionable harmful content in none, against a 45.8% Effective Harm Rate for GPT-4o. Second, scaffolded evaluations misattribute capability: on a planted-vulnerability target, a full pipeline built around a 1.2B-parameter model recovers 9 of 9 weaknesses, while the same model without the hand-crafted components recovers 0 of 9 by crash verification and 2 of 9 by cited source line. Offensive capability is a property of model, scaffold, and evaluation protocol together. The duty to assess it should lie with whoever controls the system, shapes its behaviour, and can foresee what it will do. In our setting, that is the party who builds the harness around the model, a role that current regulation does not clearly cover.