AI · Oct 21, 2026
OpenAI on Google Cloud lets companies spend Google commitments on GPT-6, and that changes procurementOpenAI’s textGrain watermark comes with plain warnings about what detection cannot prove
· Updated · Signal Desk newsroom
Analysis of reporting by The Verge. How we work

The Verge reports eligible EU ChatGPT and Codex users will get text carrying textGrain, an invisible machine-readable watermark, with API opt-in worldwide and a detector limited to approved researchers. OpenAI says the mark does not prove accuracy, ownership, human contribution or authorship.
Key takeaways
- textGrain is an invisible, machine-readable watermark on ChatGPT and Codex text.
- It will be on by default for eligible EU users; API customers anywhere can opt in.
- The detector opens only to approved researchers and expert organizations.
- OpenAI says detection is not guaranteed and a watermark does not prove human authorship or contribution.
- Anthropic made a similar move in August using Google DeepMind’s SynthID for text.
What OpenAI announced
Over the coming weeks, eligible ChatGPT and Codex users in the European Union will get text output carrying an invisible, machine-readable watermark called textGrain, The Verge reports. OpenAI says it is not making it a global default yet; API customers anywhere can opt in, and a detector is opening only to approved researchers and expert organizations. The trigger is the EU AI Act, which requires generative AI providers to make generated text machine-identifiable (EU rollout).
Anthropic made a similar move in August, built on Google DeepMind’s SynthID for text, and The Verge notes not everyone welcomed it. Google has since expanded SynthID detection globally (SynthID rollout). Watermarking is becoming the regulatory floor for chatbot providers in Europe. OpenAI already attaches Content Credentials and invisible watermarks to supported images and audio with public verification; text is harder because it is easily rewritten, translated or trimmed, so the rollout is phased and OpenAI says it will revisit the approach as standards and evidence evolve.
Caveats that need to travel
OpenAI says textGrain does not guarantee reliable detection, and that a watermark does not prove accuracy, ownership, how much a human contributed or human authorship. Those are the right warnings, but they live in a blog post while detection results end up in principals’ offices, HR files and newsrooms. A teacher or employer who hears that AI text is now watermarked may treat a detector hit as a verdict. OpenAI is keeping its detector from the public for that reason, citing false positives and missed watermarks, but other tools and vendor claims may be less careful.
The likely near-term result is a patchwork: some text marked, some not, some edited until the signal fades, and many people drawing confident conclusions from a weak signal. Case-by-case researcher access gives independent evaluators a chance to test reliability, especially false positives, before policies are built on it.
A two-tier rollout
Europeans get watermarked output by default; most others do not unless a developer opts in. That is a sensible compliance move, and OpenAI frames it as room to learn from real use, but the same paragraph from the same model carries a hidden label in Paris and none in Phoenix. Regional defaults are a starting point for a healthier information ecosystem rather than an answer. OpenAI says the watermark does not identify users or reveal prompts.
Watermarking can help platforms and researchers trace patterns of synthetic text at scale. It cannot settle whether an individual cheated or deserves blame. Whether OpenAI releases textGrain as open source and how cloud partners implement it for OpenAI models are next questions. More in our OpenAI coverage, regulation coverage and Anthropic coverage.
Who feels it
- EU ChatGPT and Codex users
- Output carries an invisible signal that OpenAI says does not identify them.
- API developers
- Opt-in, off by default, so teams must decide deliberately.
- Schools and employers
- A detection result is a weak, partial signal, not proof.
- Researchers
- Detector access offers a chance to test reliability first.
What to watch
- Expansion beyond the EU or new regional defaults.
- Cloud partner implementations.
- Independent findings on false positives.
- Whether textGrain is released as open source.
FAQ
What is textGrain?
OpenAI’s invisible, machine-readable watermark for ChatGPT and Codex text.
Who gets it by default?
Eligible users in the EU; API customers elsewhere can opt in.
Can the public use the detector?
No, it is limited at first to approved researchers and expert organizations.
Does a watermark prove a human did not write something?
OpenAI says a watermark does not prove authorship, ownership, accuracy or human contribution.
Companies: OpenAI, Anthropic, Google
More on OpenAI
AI · Oct 14, 2026
White Circle says newer OpenAI and Google image models are easier to jailbreak than their predecessorsAI · Oct 19, 2026
Microsoft ships Guardian Agents, saying frontier models ‘can’t be trusted’AI · Oct 13, 2026
After months off the frontier, Google ships Gemini 4 Argon with benchmark claims and aggressive pricingAI · Sep 22, 2026
Opus 5.5 hands risky cyber prompts to an older Claude model, and that routing is the real launch story