OpenAI to add invisible text watermarks to ChatGPT in EU

By Splitfeed EditorialLast updated Oct 6, 2026, 21:10 UTC

On October 6, 2026, OpenAI announced it would begin applying invisible text watermarks to outputs generated by ChatGPT and Codex for users located in the European Union. The feature is enabled by default for those users and is tied directly to the company's obligations under the EU AI Act, which mandates transparency measures for AI-generated content. The rollout is geographically limited, meaning users outside the EU will not have watermarking applied to their outputs by default.

The move matters because it represents one of the first concrete, large-scale implementations of AI content watermarking driven by a binding legal requirement rather than voluntary industry practice. The EU AI Act has been a reference point in global AI governance debates, and how OpenAI handles compliance will be watched by regulators and companies in other jurisdictions weighing similar rules. The central dispute here is not whether watermarking should exist but whether it can work. Ars Technica reported that text watermarking is generally unreliable and relatively easy to circumvent, a limitation that raises questions about whether the measure fulfills the spirit of the transparency requirement or simply satisfies its letter.

None of the three camps have published reactions to this story yet. Pro-AI voices would typically argue that this kind of compliance effort shows the industry acting responsibly within regulatory frameworks, and that technical imperfections in watermarking are a reason to invest in improving the technology rather than to abandon it. Anti-AI voices would typically treat the circumvention problem as evidence that watermarking is regulatory theater, providing a false sense of accountability while doing little to help people identify AI-generated content in practice. Middle-ground observers would typically call for independent auditing of the watermarking system and push for international coordination so that protections do not remain patchwork by region.

The clearest thing to watch going forward is how EU regulators assess whether OpenAI's implementation satisfies the AI Act's transparency requirements, and whether any enforcement action or guidance emerges that forces a more robust technical standard. Independent research testing the durability of the watermarks against common editing and paraphrasing would also help settle the reliability question.