OpenAI is rolling out textGrain, an invisible, machine-readable text watermark, to ChatGPT and Codex for users in the European Union, TechCrunch reported. The company will introduce the watermarking across all EU subscription tiers over the coming weeks rather than deploying it as a global default at launch. OpenAI described the regional rollout as a way to evaluate real-world feedback.
API customers worldwide can opt in to watermarked text outputs for select models starting Monday. OpenAI said this setting allows commercial clients to align with transparency obligations and adjust user experiences. The company is also working with cloud partners to support watermarking for OpenAI models hosted across their platforms over the coming weeks.
Anthropic adopted Google DeepMind's SynthID technology in August to satisfy the European Union's AI Act, a standard OpenAI said textGrain matched or exceeded. Benchmark evaluations published by OpenAI showed similar model performance between watermarked and unwatermarked text outputs.
OpenAI acknowledged that textGrain does not guarantee reliable detection. Text watermarks do not verify factual accuracy, determine text ownership, measure human contribution, or prove human authorship.
Approved researchers and expert organizations can apply starting Monday to use OpenAI's watermark detector on a case-by-case basis under the Code of Practice. OpenAI withheld the detector from the public at launch, citing risks of false positives and missed watermarks. The tool reports whether an OpenAI watermark is present without identifying the user or revealing underlying conversation prompts.
