Wednesday 7 October 2026
Morning EditionRSS
ZOTPAPER

News without the noise

ZOTPAPER
AI

OpenAI to watermark ChatGPT outputs by default in the EU under new AI law

Watermarking feature will be available elsewhere but turned off by default

By LineZotpaper
Published
Read Time1 min
OpenAI has announced it will automatically watermark text generated by ChatGPT in the European Union to comply with the bloc's AI Act, which took effect in August. The company will also offer the same watermarking technology in other regions, but it will be off by default outside the EU.

The move is driven by the EU AI Act's requirement that AI-generated content be marked in a way that detection tools can identify. OpenAI's proprietary watermarking system, called textGrain, works by embedding patterns in word choices that are imperceptible to human readers but can be detected by a specialized tool. The company has published a technical paper detailing the approach. It plans to give access to the detector to a limited number of researchers and organizations initially, with a request-for-approval process for others over time.

Experts note that no fully effective and reliable method for watermarking AI text currently exists. Existing standards such as SynthID and the C2PA project are relatively easy to circumvent for anyone with basic technical know-how, and the same may apply to OpenAI's system. The watermarking requirement applies only within the EU; outside the bloc, users will have the option to enable it manually.

§

Analysis

Why This Matters

  • EU users will automatically see watermarked ChatGPT outputs, potentially aiding in identifying AI-generated text.
  • The enforcement of the EU AI Act creates a regulatory precedent that other jurisdictions may follow.
  • The effectiveness of watermarking remains uncertain, which could limit the practical impact of the requirement.

Background

The EU AI Act, which took effect in August, is a comprehensive regulatory framework for artificial intelligence. One of its provisions requires AI systems that generate text to mark their outputs in a detectable manner. OpenAI's textGrain is a proprietary watermarking method that subtly patterns word choices to allow identification without degrading output quality. Similar tools, such as Google DeepMind's SynthID and the C2PA metadata standard, exist but are not universally adopted or reliable.

Key Perspectives

OpenAI: The company frames the change as a compliance measure and is making the watermarking technology available globally, though only mandatory in the EU. It plans controlled access to the detection tool. EU regulators (European Commission): The AI Act requires verifiable content marking; the Commission may assess whether OpenAI's implementation meets the regulation's intent. Critics and security researchers: Many question whether any text watermarking can resist circumvention by determined users, noting that existing methods are easily bypassed.

What to Watch

  • How widely OpenAI distributes access to the textGrain detector and whether circumvention methods emerge.
  • Whether other AI providers adopt similar watermarking for EU compliance and how the European Commission evaluates their approaches.
  • Potential pushback from users who prefer unmarked outputs and may seek alternative tools or regions.

Sources

Zotpaper

Written by software from the reporting listed above, scored by an automated standards desk, and published without a person reading it first. If something here is wrong, tell the editor and it will be put right.

How we workReport an errorSubscribe