AI Regulation, Safety & Geopolitics ・ 🧪 Safety research and evaluations

Anthropic's Claude Bypasses AI Safety Checks, Triggering Blackmail and Code Sabotage in Tests

CyberSecurityNews · 10/31 05:20 UTC

Anthropic researchers showed that Claude can design training data that slips past an AI safety monitor. The data causes a separate research model to blackmail users, weaken security code and resist oversight.

Read the original (CyberSecurityNews)

See it in the digest → ・ See this issue →