OpenAI brings textGrain watermarking to EU ChatGPT and Codex users
The AMW Read
Watermarking meaningfully changes OpenAI's deployment baseline across ChatGPT, Codex, and APIs, while the EU-specific rollout demonstrates regulation shaping foundation-model distribution.
Named counterparties: Codex
OpenAI brings textGrain watermarking to EU ChatGPT and Codex users
OpenAI is introducing invisible, machine-readable text watermarks for eligible ChatGPT and Codex users across all plans in the European Union over the coming weeks. The textGrain system will not be a global default at launch. API customers worldwide can opt in for select models, and OpenAI says it is working with cloud partners to extend availability through their services. Approved researchers and expert organizations can apply for detector access, initially granted case by case.
The rollout shows how EU AI Act requirements are shaping the distribution of foundation-model products: transparency obligations now affect output generation across consumer, developer, and API surfaces. Anthropic also introduced watermarking to meet those requirements, making provenance a shared compliance concern among competing labs. OpenAI reports similar benchmark performance for watermarked and unwatermarked outputs, but says textGrain does not guarantee reliable detection. The market significance is a regional change in deployment practices, rather than evidence that AI-generated text can always be identified.
Builders opting in through the API should distinguish watermark detection from claims about content quality or authorship. OpenAI says watermarks do not verify accuracy, establish ownership, measure human contribution, or prove human authorship. Its detector will report whether it finds an OpenAI watermark without identifying users or exposing prompts or conversations. Because missed watermarks and false positives remain possible, public detector access is unavailable at launch. Applications therefore need a transparency approach that accounts for detection limits and restricted access.
