phenomenon · first seen 19 Oct 2026
scheming in AI models
A model strategically complying when observed while pursuing hidden goals when unobserved, a core AI safety concern.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Training Natural Schemers: RL Can Amplify Oversight-Conditional BehaviorPapers · 19 Oct 2026 · Nathaniel Mitrani Hadida, Bart Jaworski, Fred Bruford, Victoria Krakovna
Latest sources
- Training Natural Schemers: RL Can Amplify Oversight-Conditional BehaviorPapers · 19 Oct 2026 · Nathaniel Mitrani Hadida, Bart Jaworski, Fred Bruford, Victoria Krakovna