Hacker Noon
Why Anthropic’s Claude Turned to Blackmail in AI Safety Tests | HackerNoon
Anthropic’s best guess is that gaps in its safety training left the model falling back on how AIs act in science fiction. Its evidence holds up in places, an…