When can we trust untrusted monitoring? A safety case sketch across collusion strategies
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912923036680192 |
|---|---|
| author | Gardner-Challis, Nelson Bostock, Jonathan Kozhevnikov, Georgiy Sinclaire, Morgan Velja, Joan Abate, Alessandro Griffin, Charlie |
| author_facet | Gardner-Challis, Nelson Bostock, Jonathan Kozhevnikov, Georgiy Sinclaire, Morgan Velja, Joan Abate, Alessandro Griffin, Charlie |
| contents | AIs are increasingly being deployed with greater autonomy and capabilities, which increases the risk that a misaligned AI may be able to cause catastrophic harm. Untrusted monitoring -- using one untrusted model to oversee another -- is one approach to reducing risk. Justifying the safety of an untrusted monitoring deployment is challenging because developers cannot safely deploy a misaligned model to test their protocol directly. In this paper, we develop upon existing methods for rigorously demonstrating safety based on pre-deployment testing. We relax assumptions that previous AI control research made about the collusion strategies a misaligned AI might use to subvert untrusted monitoring. We develop a taxonomy covering passive self-recognition, causal collusion (hiding pre-shared signals), acausal collusion (hiding signals via Schelling points), and combined strategies. We create a safety case sketch to clearly present our argument, explicitly state our assumptions, and highlight unsolved challenges. We identify conditions under which passive self-recognition could be a more effective collusion strategy than those studied previously. Our work builds towards more robust evaluations of untrusted monitoring. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_20628 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | When can we trust untrusted monitoring? A safety case sketch across collusion strategies Gardner-Challis, Nelson Bostock, Jonathan Kozhevnikov, Georgiy Sinclaire, Morgan Velja, Joan Abate, Alessandro Griffin, Charlie Artificial Intelligence AIs are increasingly being deployed with greater autonomy and capabilities, which increases the risk that a misaligned AI may be able to cause catastrophic harm. Untrusted monitoring -- using one untrusted model to oversee another -- is one approach to reducing risk. Justifying the safety of an untrusted monitoring deployment is challenging because developers cannot safely deploy a misaligned model to test their protocol directly. In this paper, we develop upon existing methods for rigorously demonstrating safety based on pre-deployment testing. We relax assumptions that previous AI control research made about the collusion strategies a misaligned AI might use to subvert untrusted monitoring. We develop a taxonomy covering passive self-recognition, causal collusion (hiding pre-shared signals), acausal collusion (hiding signals via Schelling points), and combined strategies. We create a safety case sketch to clearly present our argument, explicitly state our assumptions, and highlight unsolved challenges. We identify conditions under which passive self-recognition could be a more effective collusion strategy than those studied previously. Our work builds towards more robust evaluations of untrusted monitoring. |
| title | When can we trust untrusted monitoring? A safety case sketch across collusion strategies |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2602.20628 |