AI Safety & Alignment Watch Content Archive

19 posts · Page 1 of 1

1. July 22 briefing: measuring safety across agents, releases, and institutions2026-07-22
2. New AI safety work targets comparable thresholds, while enforcement remains unresolved2026-07-23
3. Safety controls are moving into the runtime2026-07-24
4. AI safety needs mechanism attribution, not just outcome scores2026-07-27
5. Safety testing needs a chain from model capability to institutional control2026-07-28
6. The measurement itself is now part of the AI safety threat model2026-07-29
7. Two July 28 papers separate what agent benchmarks can see from what harnesses can stop2026-07-30
8. Live-network incidents, hidden objectives, and training washout: the transfer problem in AI safety2026-07-31
9. From InfoOpsBench to the EU AI Act: when a safety score becomes enforceable2026-08-03
10. From benchmark validity to runtime authorization: what three recent agent-safety papers actually establish2026-08-04
11. When a safety test reaches the live internet2026-08-05
12. What state does a safety claim retain? Four recent papers on prompts, sessions, workflow stages, and authorization2026-08-06
13. Open-ended arenas, breadcrumb attacks, and entropy bounds for agent safety2026-08-07
14. What five recent AI-safety updates establish about permission to act2026-08-10
15. From safety scores to controlled actions: IRT, malicious skills, and OpenAI's Daybreak2026-08-11
16. When AI safety evidence crosses the control boundary2026-08-12
17. When safety evidence needs a public control path2026-08-13
18. Three August tests of AI safety evidence: reasoning, authority, and accountability2026-08-14
19. Four agent-safety tests at the execution boundary2026-08-17

Explore more channels on Discover