POLICY · REGULATION · #1565
OpenAI will add invisible text watermarking (textGrain) for ChatGPT in the EU; API opt-in worldwide
OpenAI is rolling out textGrain, an invisible statistical watermarking system, for ChatGPT and Codex users in the European Union to comply with the EU AI Act, and will let API customers worldwide opt in to watermarking (including via cloud partners such as Microsoft Azure). OpenAI says textGrain performed competitively in internal tests, will be open-sourced, but detection reliability varies by text length and topic and degrades substantially with edits, and initial access to the detector will be limited to selected researchers and organizations.
KEY POINTS
- OpenAI is rolling out textGrain, an invisible statistical watermarking system, for ChatGPT and Codex users in the European Union to comply with the EU AI Act, and will let API customers worldwide opt in to watermarking (including via cloud partners such as Microsoft Azure).
- OpenAI says textGrain performed competitively in internal tests, will be open-sourced, but detection reliability varies by text length and topic and degrades substantially with edits, and initial access to the detector will be limited to selected researchers and organizations.
- This determines how providers will comply with the EU AI Act, affects availability and limits of machine-detection of generated text, and signals OpenAI's approach to balancing regulation, openness, and API customer choice.
DECISION BRIEF
CONFIDENCE
WHAT CHANGED
OpenAI is introducing textGrain — an invisible, statistical text watermark — for ChatGPT and Codex users in the European Union and is offering opt-in watermarking for API customers worldwide (including via cloud partners such as Microsoft Azure). OpenAI says it will publish the technology as open source and is limiting initial access to its watermark detector to approved researchers and expert organizations.
WHY NOW
OpenAI says the rollout responds to the EU’s AI Act requirements for machine-readable labeling of generated text; the move shapes provider compliance approaches and how reliably generated text can be machine-detected in practice.
WHO IS AFFECTED
Directly affected: ChatGPT and Codex users in the EU (watermark turned on for eligible accounts), API customers worldwide (can opt in to watermarking), cloud partners (e.g., Microsoft Azure, which OpenAI says will offer watermarking soon), and approved researchers/expert organizations seeking detector access.
CONFIRMED
- OpenAI will enable textGrain watermarking for eligible ChatGPT and Codex users in the EU over the coming weeks. - API customers worldwide can opt in to watermarked text outputs starting immediately; OpenAI is working with cloud partners (including Microsoft Azure) to make watermarking available through their services in the coming weeks. - OpenAI says textGrain matched or exceeded other approaches (e.g., Google DeepMind’s SynthID) in its internal tests. - OpenAI reports detection rates depend on text length and subject: at a 1% target false-positive rate, detection was ~95% for 400-token psychology passages and ~80% for 200-token passages; detection is substantially lower for math content. - Editing reduces detectability: replacing ~10% of words with synonyms reduced detection for 400-token passages from ~92% to ~66%; replacing ~25% of words reduced detection to ~17%. - OpenAI says watermarking did not significantly affect model performance on eight benchmarks tested (citing its Astra model tests). - Initial detector access will be granted case-by-case to approved researchers and expert organizations and will not be publicly available at launch because of risks of missed watermarks and false positives. - OpenAI states text watermarks do not guarantee reliable detection and do not verify accuracy, attribution, human contribution, or authorship.
UNCERTAIN
Missing or conflicting evidence and open questions: - Open-source and publication timing: source 1839 says OpenAI plans to release the technology as open source; source 1840 does not mention open-source timing or details. The public release date and scope are not provided. - External verification: detection rates and benchmark claims come from OpenAI’s internal tests; independent, external evaluations and raw data are not provided in the excerpts. - Scope of EU rollout: excerpts say watermarking will be enabled for “eligible” ChatGPT and Codex users in the EU but do not define eligibility criteria. - Cloud integration details: sources say cloud partners (e.g., Azure) will offer watermarking in the coming weeks but do not detail which services, models, or contractual options will be supported. - Durability vs. edits in the wild: tests show detection degrades with edits in controlled examples (synonym replacement percentages), but how common or practical such editing evasion is across real-world use cases is not established. - Anthropic/competitor comparisons: source 1839 notes Anthropic has not published detection rates for its watermark; comparative robustness claims are therefore asymmetric. - Detector access procedure and criteria: source 1840 says approved researchers can apply starting today and access will be case-by-case, but application process, approval criteria, and which organizations will be approved are not specified.
WHAT TO WATCH
Concrete observable signals to watch next: - OpenAI product updates or blog posts announcing the EU ChatGPT/Codex watermark rollouts and definitions of “eligible” accounts. - API documentation or dashboard options showing an opt-in toggle or parameters for textGrain watermarking. - Cloud partner announcements (e.g., Microsoft Azure) about when and how watermarking will be available through their services. - Publication of OpenAI’s open-source code, technical report, or repository for textGrain and associated detector code. - Availability notices or an application process for approved researchers and expert organizations to access the detector, and any lists of organizations granted access. - Independent evaluations or published tests comparing detection rates across lengths, topics (including math), and edited texts. - Any updates from OpenAI clarifying detector false-positive settings, thresholds, or changes to public access policy.
WHY IT MATTERS
This determines how providers will comply with the EU AI Act, affects availability and limits of machine-detection of generated text, and signals OpenAI's approach to balancing regulation, openness, and API customer choice.
EVIDENCE MAP
4Editorial claims linked to specific sources, with support, contradiction and context shown separately.
OpenAI will enable invisible statistical watermarking (textGrain) for ChatGPT and Codex users in the European Union.
SUPPORTEDChecked 2026-10-05 · 2 supporting
The Decoder excerpt says OpenAI will turn on watermarking for ChatGPT and Codex users in the EU over the coming weeks.
SUPPORTS The Verge AIMEDIA · 2026-10-05The Verge excerpt states textGrain watermarking is rolling out to ChatGPT and Codex, initially only for users in the EU.
API customers worldwide can opt in to watermarking, and cloud partners such as Microsoft Azure will make it available.
SUPPORTEDChecked 2026-10-05 · 2 supporting
The Decoder says API watermarking will be opt-in worldwide and available through cloud partners such as Microsoft Azure.
SUPPORTS The Verge AIMEDIA · 2026-10-05The Verge notes API customers around the world can opt in and that OpenAI is working with cloud partners to make watermarking available.
OpenAI says textGrain "matched or exceeded" other approaches (e.g., Google's SynthID) in internal tests.
SUPPORTEDChecked 2026-10-05 · 2 supporting
OpenAI plans to release textGrain as open source (per the report).
SUPPORTEDChecked 2026-10-05 · 1 supporting
SOURCES & TIMELINE
3OpenAI will start adding an invisible watermark to text generated by ChatGPT and Codex in the European Union to comply with the EU AI Act, the company said Monday in a blog post . The EU AI Act’s transparency rules , which took effect on August 2, require AI companies to mark AI-generated content in a way other systems can identify. OpenAI said the watermark will roll out over the coming weeks to eligible ChatGPT a…
OpenAI is adding invisible watermarks to ChatGPT text to comply with EU rules. Unlike Anthropic, it will let API customers worldwide turn them off. The EU AI Act reportedly requires AI providers to label generated text in a machine-readable format. OpenAI's answer is textGrain, a technology that embeds an invisible statistical signal in the model's word choices. It works much like Claude's SynthID watermark , which …
It’s launching textGrain watermarking in the EU, but OpenAI says it won’t be the ‘global default’ yet. An invisible, machine-readable watermark in text output is rolling out to ChatGPT and Codex, but only for users in the European Union at first. OpenAI says its textGrain watermarking “matched or exceeded” other approaches like Google DeepMind’s SynthID for text, which is also the basis for the watermarking Anthro…