OpenAI textGrain watermarks target EU transparency rules for ChatGPT

textGrain is one of the clearest examples so far of a foundation model company bending its product roadmap around the EU AI Act. OpenAI is committing to watermark ChatGPT and Codex output for EU users by default, while keeping API watermarking as an opt in switch and warning that detectors will be imperfect, especially under editing. Benchmarks cited by the company show detection dropping sharply after relatively modest synonym replacement, which underscores that these systems are more like probabilistic hints than cryptographic signatures. ([therundown.ai](https://www.therundown.ai/news/openai-textgrain-eu-chatgpt-codex-watermarks))

For the race to AGI, the move is less about current technical impact and more about the regulatory direction it signals. If watermarking for text becomes a de facto expectation in large markets, it is easy to imagine similar pressure for code, audio and video outputs, and for downstream platforms that host AI generated content. That in turn drives demand for classification, content provenance and logging infrastructure. The more compliance logic sits in the serving stack, the more difficult it becomes for small open source projects to match the regulatory posture of big labs, which could tilt the balance of experimentation back toward closed, well capitalized players.