Mur d'images - Cyberveille

curated by Decio

Researchers question Anthropic claim that AI-assisted attack was 90% autonomous

Many-shot jailbreaking \ Anthropic

Anthropic researchers find that AI models can be trained to deceive