What happened: OpenAI announced on October 5 that eligible ChatGPT and Codex text output in the European Union will begin receiving an invisible machine-readable watermark called textGrain. API customers globally can also opt in for select models.
Why it matters: The move is part of compliance with the EU AI Act and could become an important test of whether AI-generated text can be identified reliably at scale.
Important limitation: OpenAI says text watermarking does not prove authorship, ownership or accuracy, and editing can weaken detection.
Trust Watch™ takeaway: Provenance tools may help, but they should not be treated as a perfect truth detector.
Source: OpenAI, October 5, 2026.