OpenAI adds text watermarking to ChatGPT and Codex, initially only for EU users

OpenAI is adding invisible, machine-readable watermarks to text outputs from ChatGPT and Codex, initially only for EU users, rolling out to eligible users across all plans in the EU in the coming weeks, and not setting it as a global default for now. OpenAI says its textGrain watermark performs comparably to or better than approaches like Google DeepMind's SynthID in evaluations, and Anthropic's watermark announced in August is also based on SynthID; both aim to meet the requirements of the EU AI Act. API customers can opt in to watermarked outputs on some models starting today, and OpenAI also plans to offer the capability with cloud partners in the coming weeks. Approved researchers and expert institutions can apply to use the detection tool starting today, with initial approvals on a case-by-case basis. The tool only reports whether an OpenAI watermark is detected; it does not identify users or reveal prompts or conversations. OpenAI also cautions that textGrain does not guarantee reliable detection, and that the watermark cannot verify content accuracy, determine text authorship, measure the proportion of human contribution, or prove that text was written by a human.