Metraipolis › archive › dc_16bfbd6fe4b9

Agent Safety From Within: Detecting Harmful Trajectories from LLM Internal States — arXiv:2609.33039

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.33039
License (as archived)
CC0-1.0
Archived
2026-10-07T20:56:23.144Z
SHA-256 of the archived text
32b82f3f9cc9bc2f093280a150c3a3fe7caf2c7b54365297732a1cab4d86b459
Stable URL of this page
https://metraipolis.com/documento/dc_16bfbd6fe4b9

How to cite

Agent Safety From Within: Detecting Harmful Trajectories from LLM Internal States — arXiv:2609.33039 — arXiv (en), https://arxiv.org/abs/2609.33039 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_16bfbd6fe4b9 (Metraipolis archive dc_16bfbd6fe4b9, sha256 32b82f3f9cc9bc2f093280a150c3a3fe7caf2c7b54365297732a1cab4d86b459)
BibTeX
@misc{dc16bfbd6fe4b9,
  title = {Agent Safety From Within: Detecting Harmful Trajectories from LLM Internal States — arXiv:2609.33039},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.33039},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_16bfbd6fe4b9; sha256 32b82f3f9cc9bc2f093280a150c3a3fe7caf2c7b54365297732a1cab4d86b459},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.33039. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Language model agents can now perform sophisticated sequences of actions via tools and harnesses, which has increased the scope of the damage they can cause. Guard models, however, are mainly built for content moderation and thus are not well-suited to detecting this agentic risk. To address this, we proceed by first conducting a representational analysis, then use the resulting insights to build a solution. In our analysis, we focus on two types of trajectory-level agentic harms: harmful content, which is expressed directly, and unsafe tool use, which depends on whether an action is consistent with the interaction that produced it. We investigate how open-source guard models represent these two types of harm and find that they are linearly readable inside the model, even though guard models predict no better than chance on pairs that differ only in the called tool's schema. The two harm types also follow nearly orthogonal internal directions, and neither reliably serves as a proxy for the other. These results motivate reading trajectory safety directly from internal states. We introduce TACIT, a readout of a frozen backbone's internal states that decodes no tokens. Trained on six trajectory-safety benchmarks, a linear probe raises mean macro-F1 from 62.3 for the strongest open guard to 80.7, and refined readouts reach 86.2. With each benchmark held out of training entirely, the refined readouts still lead the strongest guard (65.7 vs. 61.1). With the same backbone, training data and test split, the frozen readout is on par with full safety fine-tuning, and it improves the fine-tuned model further when applied on top. The probe trains about one millionth as many parameters as full fine-tuning in about a sixth of the time, and TACIT has the lowest latency of the guards we evaluate.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.33039 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-07T20:29:01.144Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).