OpenAI adds optional text watermark API feature to meet EU AI Act rules

OpenAI will add a text watermark to ChatGPT outputs for users in the European Union. The goal is to comply with the bloc’s rules on text provenance, which require AI-generated content to be marked in a way machines can detect. What OpenAI is actually rolling out The change is tied to Article 50 of the EU AI Act. That provision covers transparency, and it calls for generative AI outputs to carry machine-readable marking. Official OpenAI documentation published in early October 2026 lays out the broader setup. Images and audio will carry two signals: C2PA metadata and SynthID watermarks. C2PA is a metadata standard that attaches origin information to a file. SynthID is an imperceptible watermark developed by Google DeepMind, which OpenAI has licensed to strengthen its media verification tools. On the text side, the documentation describes new watermarking features that can be switched on within API project and organization settings. Developers building on OpenAI’s platform also get access to a Content Provenance API for eligible use cases. The documented approach treats text watermarking as something to activate, and it is not a default feature applied to every ChatGPT output. OpenAI...