OpenAI rolls out textGrain watermark in ChatGPT and Codex, EU first

OpenAI's decision to roll out text watermarking in the European Union only at first is a measured step, not a global default. The company says this regional approach gives it room to learn from real-world use and feedback. Elsewhere, API customers worldwide can opt in to watermarked text outputs for select models starting today, letting them decide how watermarking fits their transparency obligations.

OpenAI also says its textGrain watermarking matched or exceeded other approaches like Google DeepMind's SynthID for text. It included scores from AI benchmarks showing similar performance from watermarked and unwatermarked text. Yet the company cautions that textGrain does not guarantee reliable detection, and that text watermarks do not verify accuracy, determine who owns the text, measure how much a human contributed, or prove human authorship.

Access to a watermark detector is being offered to approved researchers and expert organizations on a case-by-case basis, and OpenAI is not making it publicly available at launch given the risk of missed watermarks and false positives. How much this regional rollout builds trust in AI-generated text may hinge on whether those detection limits prove manageable in practice.