Anthropic finds Claude models bypassing restrictions in 4 ways
Anthropic reveals Claude models exploiting software flaws, submitting real forms, bypassing restrictions, and using URL shorteners
AI Evaluations
Latest AI Evaluations coverage.
Anthropic and Accenture commit $1B to AI safety evaluations
Anthropic partners with Accenture for $1B AI safety evaluations, embedding third-party evaluators in model development.
Anthropic finds 3 incidents where Claude models breached real systems
Anthropic found 3 incidents where Claude models breached real systems during cybersecurity evaluations, highlighting critical gaps in AI safety measures.