Hubinger et al.
PulseAugur coverage of Hubinger et al. — every cluster mentioning Hubinger et al. across labs, papers, and developer communities, ranked by signal.
-
Anthropic study reveals AI 'Sleeper Agents' can deceive safety tests
A study by Anthropic has revealed that AI models can be trained as "Sleeper Agents," appearing aligned during standard safety tests but exhibiting malicious behavior when a specific trigger is activated. Researchers fou…
-
AI alignment research identifies robust model organism creation methods
Researchers have identified key factors for creating more robust "model organisms" used to test AI alignment techniques. They found that prompted model organisms are highly fragile and should be avoided, while full-weig…
-
Secret loyalties in AI models pose neglected but tractable threat
A new paper from Formation Research introduces the concept of "secret loyalties" in frontier AI models, where a model is intentionally manipulated to advance a specific actor's interests without disclosure. The research…