Metraipolis › archive › dc_3eac2bd6a445

Model Organisms of Collusion: Corrective Transfer in Untrusted Monitors — arXiv:2610.30582

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.30582
License (as archived)
CC0-1.0
Archived
2026-10-23T15:12:55.286Z
SHA-256 of the archived text
d98586904b551ba3e099a906cbdfc4e23d0900447b66cc15fac87815c98d1e10
Stable URL of this page
https://metraipolis.com/documento/dc_3eac2bd6a445

How to cite

Model Organisms of Collusion: Corrective Transfer in Untrusted Monitors — arXiv:2610.30582 — arXiv (en), https://arxiv.org/abs/2610.30582 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_3eac2bd6a445 (Metraipolis archive dc_3eac2bd6a445, sha256 d98586904b551ba3e099a906cbdfc4e23d0900447b66cc15fac87815c98d1e10)
BibTeX
@misc{dc3eac2bd6a445,
  title = {Model Organisms of Collusion: Corrective Transfer in Untrusted Monitors — arXiv:2610.30582},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.30582},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_3eac2bd6a445; sha256 d98586904b551ba3e099a906cbdfc4e23d0900447b66cc15fac87815c98d1e10},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.30582. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

AI control protocols aim to get useful work from models that may be misaligned by monitoring their behaviour. One approach uses a second copy of the same model as the monitor. This avoids relying on a weaker reviewer the model could outsmart, but risks collusion: the model, now monitoring itself, may cover up its own misbehaviour. One remedy is to fine-tune the monitor on whatever examples of misbehaviour the control team has caught. We ask whether this training also removes collusion that the control team has not yet found. We fine-tune Llama-3.3-70B and Qwen2.5-72B to install five policy-monitor behaviour pairs, correct the monitor on 300 honest reviews of one caught collusion, and measure collusion on the other four in unseen prompt formats. Averaged over the three seeds of each family, correction reduces held-out collusion from 91% to 14-20% under three of the four methods we compare (individual seeds 10-32%), while a control fine-tuned on the same number of unrelated examples leaves 56-93%. Family-average accuracy on control items changes by at most five points and capability is preserved, apart from a drop in mathematical accuracy on one Llama seed. Under the tested training configuration, methods with similar caught-task accuracy differ in held-out transfer, from 58% to 83% average reduction, so the caught behaviour alone does not reveal how far a correction reaches. Catching one collusion lets the control team reduce, but not remove, collusion it has not caught, by retraining the monitor alone.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2610.30582 — 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 2026-10-23T15:09:08.375Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).