OpenAI is implementing optional text watermarking features for ChatGPT users in the European Union to comply with Article 50 of the EU AI Act. This regulation mandates that generative AI content must feature machine-readable markings to ensure transparency.
The new tools are available within API project and organization settings rather than being applied by default. Developers can also utilize a Content Provenance API. For images and audio, OpenAI is integrating C2PA metadata and SynthID watermarks licensed from Google DeepMind.
These measures follow OpenAI's commitment to the EU Code of Practice on Transparency of AI-Generated Content signed in June 2026. The company previously delayed a broad release of its internal watermarking prototype, citing concerns regarding text robustness and potential bias against non-native English speakers.
The rules under Article 50 took effect on August 2, 2026, with a transition deadline for existing systems set for December 2, 2026. Competitors such as Google and Anthropic have already introduced similar watermarking capabilities for their respective AI models.
