OpenAI will add a text watermark to ChatGPT outputs for EU users. The move meets Article 50 of the EU AI Act. Article 50 requires machine-readable marking for AI-generated content. Images and audio will use C2PA metadata and SynthID watermarks. Text watermarking is opt-in via API settings. Developers get a Content Provenance API for eligible cases. OpenAI delayed due to robustness and equity concerns. Google and Anthropic already offer similar text watermarks. The EU AI Act transition deadline is December 2, 2026. Regulators may judge opt-in tools versus default watermarks.
What changed
Text watermarking was not available for ChatGPT outputs in the EU.
What it unlocks
Marking AI-generated text in a machine-readable way for EU compliance.
- 99.9% effective prototype
- August 2, 2026 Article 50 effective date
- December 2, 2026 transition deadline
What you need to act on it
- EU user base
- API project or organization settings activation
- eligible use case for Content Provenance API
Sources