Metraipolis › archive › dc_79dc3a97a185

Activation Probes Surface Code-Security Signals that the Model's Output Misses — arXiv:2608.09643

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.09643
License (as archived)
CC0-1.0
Archived
2026-10-19T08:58:25.019Z
SHA-256 of the archived text
e8d8a887d7acb713f9b14f329c9a8acd813201ae72cc53cdac43c618d525ab4c
Stable URL of this page
https://metraipolis.com/documento/dc_79dc3a97a185

How to cite

Activation Probes Surface Code-Security Signals that the Model's Output Misses — arXiv:2608.09643 — arXiv (en), https://arxiv.org/abs/2608.09643 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_79dc3a97a185 (Metraipolis archive dc_79dc3a97a185, sha256 e8d8a887d7acb713f9b14f329c9a8acd813201ae72cc53cdac43c618d525ab4c)
BibTeX
@misc{dc79dc3a97a185,
  title = {Activation Probes Surface Code-Security Signals that the Model's Output Misses — arXiv:2608.09643},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.09643},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_79dc3a97a185; sha256 e8d8a887d7acb713f9b14f329c9a8acd813201ae72cc53cdac43c618d525ab4c},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.09643. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

AI coding agents now write a growing share of production code, and human security review does not scale at the rate code is generated. The agents in widest use are closed-weight, so a deploying team cannot read their internals. It can instead run an open-weight model as a reviewer over the agent's output. That reviewer's activations are readable. We ask whether reading those activations recovers a security signal that simply asking the same reviewer misses. We fit a single linear probe per model on a corpus of paired vulnerable-and-fixed Python functions, then test it without retraining on real disclosed vulnerabilities whose weakness type the probe never saw in training, across five open-weight reviewer models. On the vulnerabilities fixed by changing a single function, the probe scores the vulnerable function above its fix on 61-67% of cases for every model, beating the 50% chance line. It also beats the same model's prompted YES/NO win-rate read from its logits, under every prompt we try. Asking the model for a written verdict, even with chain-of-thought, returns the same answer on the vulnerable and fixed function most of the time and so cannot tell them apart. Model activations carry a code-security signal that prompting the same model misses.

Tema degli esploratori: fiducia digitale e crittografia (retro-catalogo arXiv: annunci del 2026-08-11 in cs.CR).
Categorie arXiv: Cryptography and Security (cs.CR) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2608.09643 — 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 2026-10-19T08:54:24.975Z dalla pagina https://arxiv.org/catchup/cs.CR/2026-08-11?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).