AI developer Anthropic announced Monday that all new Claude models launched on or after August 2, 2026, will automatically embed invisible, machine-readable watermarks directly into generated text.
The move comes as Anthropic signs onto transparency commitments under Article 50 of the European Union's AI Act. According to an updated support disclosure from Anthropic, the company will apply the watermarking technology globally across all user access points—including the main web interface, developer APIs, coding applications, and major cloud hosting partners like AWS, Google Cloud, and Microsoft Foundry.
Company engineers explain that the watermarks function at the model level by weaving subtle, statistical patterns into word selection without altering the meaning, readability, or formatting of the output. The embedded signal travels with the text when copied and pasted into other applications, and Anthropic states the mark is capable of surviving light to moderate editing. Additionally, generated image and vector files will carry digitally signed metadata following open C2PA standards.
🚨 JUST IN: Claude models will now have invisible watermarks embedded in ALL text, and ALL metadata attached to files… pic.twitter.com/TqBF55dRoK
— NIK (@ns123abc) August 10, 2026
Anthropic stated it is actively building verification tools to allow businesses, educators, and third parties to detect watermarked content. However, the company emphasized that finding a watermark indicates a file or passage was processed by Claude, not that it was entirely written by AI, noting users frequently rely on models to proofread or format human-authored drafts.
Anthropic confirmed it is also working to retrofit older models launched prior to August 2026 during a standard regulatory transition period.
LISTEN on the Audacy App
Tell your Smart Speaker to "PLAY 1080 KRLD"
Sign Up to receive our KRLD Insider Newsletter for more news
Follow us on Facebook | Twitter | Instagram | YouTube
