High impact

Our approach to EU text provenance rules

What happened

OpenAI announced a phased rollout of text watermarking, called textGrain, to meet EU AI Act requirements that generated text be machine‑readable. Starting now, API customers can opt in, but the feature is off by default and will first apply to eligible ChatGPT and Codex outputs in the EU. The invisible watermark embeds a statistical signal in word choices, and a detector looks for this signal, though detection varies with text length, subject matter, and editing. OpenAI is opening limited access to the detector for approved researchers and expert organisations, and plans to release the technology as open source. The company also continues to offer public tools for image and audio provenance.

Why it matters

For everyday readers, this means future AI‑generated text may be identifiable, helping to spot synthetic content and reduce misinformation. It also signals that regulators are shaping how AI providers disclose model output.

Reality check

An automated, sceptical read of the source

Mostly marketingIncrementalNotableGenuine breakthrough

Incremental. The rollout adds a compliance feature with known limitations rather than a breakthrough in detection accuracy.

Who it affectsDevelopersBusinessesResearchers

What to watchWatch for broader detector access and the open‑source release of textGrain, which will show how the technology evolves and how widely it is adopted.

Do you agree with this rating?Agree Disagree

More from OpenAI

Introducing ChatGPT WalletOpenAI · 13 Nov 2026Shield helps technicians hand off legacy portal work with OpenAIOpenAI · 13 Nov 2026OpenAI launches ChatGPT Wallet, AI agents pay for purchaseOpenAI · 13 Nov 2026OpenAI disbands its robotics team as its general models start running robotsOpenAI · 13 Nov 2026Yang-Mills, OpenAI proof, and the mass gap problemOpenAI · 13 Nov 2026

This page was collected from openai.com and summarised automatically by AIPulsee's pipeline. See how it runs.