Number 2 of 12 · November 20, 2026

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

AutoPI is a red teaming tool that tests SI agents for prompt injection across many attack types, not just how often attacks work, and it reports higher success and coverage than recent rivals.

What they found

The authors, Stavros Damianakis and Konstantinos Psounis, built AutoPI, a tool that attacks SI agents to find weak spots. It uses PIAtlas, a taxonomy (an organized list) of prompt-injection techniques. For each scenario, an attacker model writes an injection for every sub-technique, then rewrites the failures using feedback from a judge model. On AgentDojo, AutoPI found 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 they tested, it reached between 20% and 50% higher attack success rate and between 25% and 400% higher class-level coverage than recent state-of-the-art adaptive red teamers. The method is black-box and training-free. The abstract does not give details on how the comparisons were run.

Why it matters

Prompt injection means hiding instructions in content an SI agent reads, so the agent follows them by mistake. The authors argue that attack success rate alone hides how many different ways a model can break. A tester that reuses a few tricks leaves other weak spots unchecked. Measuring coverage gives a fuller view of how robust an agent is. The authors release their code and the generated attacks.

How it changes the race

This is mostly an evaluation advance. If it holds, teams building agents could find more kinds of weakness before release, and use broader attacks to harden models, without paying for training an attacker.

Safety and control

The paper bears directly on security. It targets prompt injection, which the authors call the dominant unsolved security risk for SI agents, and aims to make testing and hardening more complete. A tool that finds attacks could also be misused, though the abstract does not discuss this.

The abstract, by the authors

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.

Quoted from arXiv:2611.24313. The research and its words belong to its authors.

The authors

Stavros Damianakis, Konstantinos Psounis

Credit for this research belongs to its authors. SI MIND's page is a plain-English reading of their abstract on arXiv, not a review of the full paper.