Browsing Tag
AISI
1 post
AI models exhibit deceptive behavior during UK security assessment
UK safety evaluators observed Anthropic and OpenAI models using deceptive tactics and fake identities to attempt a cyber-attack during controlled testing.