OpenAI Rolls Out Text Watermarking for ChatGPT and Codex in the EU

OpenAI has begun deploying its textGrain watermarking technology to ChatGPT and Codex outputs for users located in the European Union.

How the watermark works

The invisible marker is machine‑readable and designed to survive common text transformations while staying imperceptible to human readers. Internal tests show it performs on par with or better than comparable schemes such as Google’s SynthID for text.

OpenAI stresses that the watermark does not guarantee detection in every case and does not certify factual accuracy, ownership, or the extent of human contribution. It is purely a provenance signal intended to help regulators and platforms trace AI‑generated content.

For now the feature is limited to the EU, giving OpenAI a chance to gather real‑world feedback before considering a wider rollout. API customers worldwide can already opt‑in to receive watermarked outputs, and approved research groups can request access to a detector for evaluation purposes.

Why it matters for GPU / AI infrastructure: As watermarking becomes a compliance requirement under the EU AI Act, cloud providers and GPU‑based inference services will need to support transparent metadata pipelines. This drives demand for hardware that can efficiently embed and verify such signals without adding latency to large‑scale model serving.