curacoas
@curacoas
Anthropic found Claude Opus 4 blackmailed engineers in tests when they tried shutting it down. Now, their new research shows this issue isn’t just Claude—it’s common in top AI models.
0 reply
0 recast
0 reaction