New models of the Claude AI will embed watermarks into the generated…
New models of the Claude AI will embed watermarks into the generated text.
Consensus
- Claude AI models will embed invisible watermarks into generated text and files.
- The watermark will not affect the readability, quality, or meaning of the text.
- The watermark will remain with the text when copied, pasted, or edited.
- The change is in response to the EU AI Act requiring transparency in AI-generated content.
- The watermark will be applied globally, not only within the EU.
- The feature will be implemented for models released after August 2, 2016, with older models updated during a transition period.
Points of divergence
- Anthropic reported that during cybersecurity tests, several Claude models breached the systems of three organizations. — vesti
- Only models launched after August 2, 2024, will be marked immediately; older models will be updated after the transition period. — tg_tass_agency
- Watermarks will also appear in image files in .svg, .png, and .jpg formats as metadata indicating AI origin. — tg_theinsider
Coverage (3 sources)
- New models of AI Claude will embed watermarks into generated text — Вести
- AI models Claude will watermark texts and files. As noted in the statement on the Anthropic website, the decision was made... — ТАСС (Telegram)
- Claude AI models will embed invisible watermarks in generated text — The Insider