Back to AI News
arxiv

AutoPI: Taxonomy-Guided Adaptive Red Teaming for Prompt Injection in LLM Agents

AutoPI is a black-box, training-free adaptive red teamer that uses PIAtlas, a taxonomy of prompt-injection techniques, both to guide its search and to measure coverage, rather than optimizing attack success rate alone. For each scenario, an attacker LLM writes an injection for every sub-technique in the taxonomy and then reworks the failures using feedback from an LLM judge. The authors evaluate AutoPI on AgentDojo and IPI Arena against open-weight and proprietary target models, reporting technique coverage alongside ASR. Against Claude Haiku 5.5, described as the most robust target evaluated, AutoPI reports 20% to 50% higher ASR and 25% to 400% higher class-level coverage than recent state-of-the-art adaptive red teamers, with code and generated attacks released.

AI Brief

It offers a coverage-aware, training-free method for red-teaming prompt injection in LLM agents, a dominant unsolved security risk, with released code and attacks.

arXiv:2611.24313v1 Announce Type: cross Abstract: Prompt injection is the dominant unsolved security risk for LLM agents, and the community now agrees that robustness must be measured against adaptive attackers rather than fixed benchmark injections. Many adaptive red teamers exist, and most are built to maximize attack success rate (ASR). Yet ASR counts how often a target breaks, not in how many different ways. A red teamer can achieve a high ASR by reusing a few techniques that work, leaving the others untested and absent from the attacks used to harden models. We present AutoPI, an adaptive red teamer designed to find many different ways to break a target, not just to break it often. AutoPI uses PIAtlas, a taxonomy of prompt-injection techniques, both to guide its search and to measure coverage. For each scenario, an attacker LLM writes an injection for every sub-technique in the taxonomy and then reworks the ones that fail using feedback from an LLM judge. The method is black-box and training-free, it plugs into existing benchmarks in place of their fixed attack templates, and it stays up to date as the taxonomy evolves. We evaluate AutoPI on AgentDojo and IPI Arena against open-weight and proprietary target models. Alongside ASR, we report technique coverage, the fraction of taxonomy classes, techniques, and sub-techniques that yield at least one successful injection. On AgentDojo, AutoPI finds a successful injection against Qwen3.6-35B in all six classes of the taxonomy and for more than 90% of its techniques. Against Claude Haiku 5.5, the most robust target we evaluate, it achieves between 20% and 50% higher ASR and between 25% and 400% higher class-level coverage than recent state-of-the-art adaptive red teamers, while requiring no training. We release our code and the generated attacks to support reproducible, coverage-aware evaluation.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!