Anthropic Risk Report Reveals Claude's R&D Acceleration and Safety Lapses

Anthropic published a risk report showing Claude now writes majority of Anthropic's production code, internal R&D is accelerating, safety evaluations are saturating, and they've lowered confidence in risk assessments. Disclosed safety lapses like chain-of-thought leakage and model refusal going unnoticed. This challenges the narrative of full control and reveals real operational risks at frontier labs. Anthropic is also implementing global watermarking for Claude via C2PA and embedded text watermarks, setting a new transparency standard.