OpenAI will watermark ChatGPT text by default, but only in the EU
What's the deal? OpenAI says it will add an invisible watermark to eligible ChatGPT and Codex text for users in the European Union "over the coming weeks", across all plans. Outside the EU it will not be on by default: API customers worldwide can opt in for select models, but it stays off unless they switch it on. OpenAI announced the plan on 5 October in a post titled Our approach to EU text provenance rules.
How it works: The system, called textGrain, nudges the model's word choices to leave a statistical signal that someone holding the key can detect. OpenAI published a technical report alongside the post. It says the signal survives copy and paste, but editing, translation, short passages and constrained outputs such as maths weaken it, and that a missing watermark does not prove a human wrote the text. Access to the detector starts with approved researchers and expert organisations only.
Why now? The EU AI Act requires providers of generative AI to mark generated content in a machine-readable, detectable way. That obligation has applied since 2 August 2026, and according to ActuIA, systems already on the market before then have until 2 December 2026 to comply.
The contrast: AnthropicDealroom has a profile for this one. Try Dealroom → introduced text watermarking for its models in August and applies it globally, Ars Technica reports. OpenAI is making it the default only where regulators require it.
The signal: This is one of the first times European AI rules have produced a visibly different version of a mainstream AI product for EU users. Whether other labs follow OpenAI's EU-only default or Anthropic's global one will show how far the "Brussels effect" reaches in AI. That reading is our analysis rather than anything OpenAI has said.
Image credit: OpenAI
