Can Anthropic’s invisible watermarks curb ‘AI slop’? Researchers remain sceptical

August 2026

Anthropic announced text generated by Claude models launched on or after August 2, 2026 will carry invisible statistical watermarks. Images use C2PA metadata signatures instead.

EU AI Act Article 50 requires machine-readable marking of generative outputs as of August 2, 2026. Anthropic applies watermarks globally.

Researchers note watermarks reveal little about how models were used; short, paraphrased, or rewritten text may lose the signal. Detection is planned but not yet publicly available.