Anthropic says it will watermark text produced by its models, including Claude, as part of complying with European transparency rules tied to the EU AI Act. Reporting on the company’s updated support guidance notes that models released after August 2, 2026 automatically include watermarking for generated text and files, with file provenance using the C2PA open standard. Anthropic says the text watermark is embedded at the model level, so it should appear across Claude surfaces (chat, API, Claude Code, and related products) and can travel when users copy and paste—sometimes surviving light edits. Other major labs have also signaled support for the EU transparency code. Separately, some Claude users publicly worried the watermarks could expose AI-assisted school or workplace writing. For enterprises, the practical effect is dual: easier automated detection of AI-origin text for compliance teams, and a new honesty layer when content moves between tools. Exact robustness against heavy rewriting is still an open question Anthropic has been pressed to clarify.
Source: https://techcrunch.com/2026/08/11/anthropic-says-it-will-watermark-text-generated-by-its-ai-models/