Reliability Theory for AI Control
Not on the map. Why?
Abstract (arXiv)
Reliability theory gives a mature language for layered systems, but its formal tools are not yet standard in frontier AI control. We apply them to Google DeepMind's defenses against rogue deployment. The same control stack can have cubic, quadratic, or linear rare-failure suppression depending on its failure domains. Birnbaum importance identifies which component improvements buy the most nominal reliability, while prevention changes the population on which recovery is demanded. These results give concrete guidance about what to separate, improve, measure, and test.
Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Builds on 10 of its 30 references that are in the map. First posted on arXiv 22 Sep 2026.
Builds on
10 of 30 references.
References · 10
- 2025Ctrl-Z: Controlling AI Agents via Resamplingcited by 40
- 2025Evaluating Control Protocols for Untrusted AI Agentscited by 9
- 2024Assurance of AI Systems From a Dependability Perspectivecited by 5
- 2026Ensemble Monitoring for AI Control: Diverse Signals Outweigh More Computecited by 5
- 2025Factor(T,U): Factored Cognition Strengthens Monitoring of Untrusted AIcited by 4
- 2025Adapting Probabilistic Risk Assessment for AIcited by 3
- 2026N-Version Programming with Coding Agentscited by 2
- 2026GDM AI Control Roadmapcited by 2
- 2026Retrying vs Resampling in AI Controlcited by 1
- 2026Attack Selection in Agentic AI Control Evaluations Meaningfully Decreases Safetycited by 1
Built on by
Citing papers per year: 2026: 1.
1 paper.