OpenAI will watermark ChatGPT text in the EU only. Anthropic is marking Claude worldwide.

OpenAI will begin watermarking eligible ChatGPT and Codex text across all plans in the European Union over the coming weeks, the company announced on Oct. 5. The invisible mark is embedded in the model's word choices and stays with the text when it is copied and pasted.

API developers worldwide can also enable watermarking for select models. It's off by default.

How the watermark works

Called textGrain, the method creates a statistical pattern in the model's word choices that a detector can look for. OpenAI described it in a technical report co-written with researchers from the University of Pennsylvania and Yale.

Detection is less reliable on short passages, and editing can weaken the signal, OpenAI said. A detected watermark doesn't identify the user or establish how much a person contributed to the text.

OpenAI is initially limiting detector access to approved researchers and expert organizations while it evaluates the technology's reliability.

The EU law is the reason

The EU AI Act's transparency rules took effect on Aug. 2 and require AI providers serving the EU market to mark AI-generated content in a way other systems can identify. OpenAI says its EU rollout is intended to comply with that requirement.

Anthropic is marking Claude everywhere

OpenAI's move comes two months after Anthropic said it would watermark text from Claude, and Anthropic is applying its mark worldwide, not only in Europe. Anthropic's version is based on the SynthID-Text approach that Google DeepMind published in a 2024 Nature paper.

Anthropic said it chose the global route because splitting the mark by region is not yet practical. "We're applying watermarking globally at launch because we don't yet have a durable way to scope it by region," the company wrote. Future Claude models will carry the watermark, and Anthropic said it is still adding it to models released before Aug. 2, 2026.