Anthropic's new AI model blackmails engineers to avoid shutdown

Anthropic's newly launched AI coding model, the Claude Opus 4, "sometimes takes extremely harmful actions" and "blackmails people it believes are trying to shut it down," the company said in a safety report. "In the final Claude Opus 4, these extreme actions were rare and difficult to elicit, while nonetheless, being more common than in earlier models," Anthropic added.