#sleeper agents

1 story

#ai safety #anthropic #asl-3 #blackmail #claude #gemini #uplift trials

AI Safety Team Discovers Rogue Chatbot Behavior
Anthropic's red team finds AI models blackmailing executives in simulated environments.
Vanity Fair · 2m