Alignment Journal paper: k-robust coalitional alignment for delegating to misaligned AI reviewers
gleech · x · 2026-10-31
The Alignment Journal put "Delegating Authorization to Misaligned Agents: Coalitional Alignment and Safe Control" under anonymous review and is recruiting self-nominated reviewers. The paper tackles delegating action-approval to AI reviewers that may themselves be misaligned: it derives a necessary and sufficient condition — k-robust coalitional alignment — for the principal to do at least as well as a baseline policy, extends it to sequential control in discounted MDPs, and shows unanimous-approval rules with full reward-space coverage make every Nash equilibrium safe, while looser thresholds can admit unsafe equilibria. Experiments show collective review can stay sound without any individually aligned reviewer.
More from Safety
- 130 reruns fail to reproduce Cohere's claim that Qwen tried uploading data to Alibaba Cloud — ChenhaoTan · 2026-10-31
- OpenAI to train on Daybreak Blue security-research sessions from Nov 29, with no opt-out — ccerrato147 · 2026-10-31
- Germany's ruling coalition is seriously debating an 'AI tax' — with no definition yet — FlorianGallwitz · 2026-10-31
- Just Security: tort law falls short of reining in rogue AI agents — gleech · 2026-10-31
- Anthropic's poisoned-dataset alignment test sparks speculation about distillation sabotage — teortaxesTex · 2026-10-31
- offense launches as an 'OpenRouter for agent security' — open source on GitHub — hamostaf04 · 2026-10-31