# Regulation Forces AI Text Watermarking

EU AI Act transparency rules push frontier labs to deploy text watermarking region by region, then extend it as an opt-in elsewhere. Because current watermarks fail under light paraphrasing, expect more vendor rollouts, published robustness attacks, and regulatory arguments over whether weak watermarks satisfy the disclosure duty, with compliance costs and detection tools forming a market of their own.

- Conviction: 30 / 100 (weakening)
- Horizon: Emerging (watchlist)
- Tracking since: 2026-10-06T00:00:00.000Z
- Last updated: 2026-10-12T14:07:13.413Z
- Canonical: https://polylog.news/ai/trends/ai-text-provenance-mandates
- Publisher: Polylog
- Affected regions: Europe, Global

## Recent score history

- 2026-10-11: 32
- 2026-10-12: 30

## Recent evidence

- [confirms] OpenAI will watermark ChatGPT and Codex text in the European Union to comply with the AI Act (2026-10-07): To comply with the EU AI Act, OpenAI will watermark ChatGPT and Codex text in the EU with a statistical mark called textGrain. It was detected in about 95% of 400-token passages in OpenAI's tests, but swapping one word in ten for a synonym cut detection to about 66%. This matches the thesis on both points: a region-by-region rollout forced by regulation, and a weak mark that invites argument over whether it satisfies the disclosure duty.
- [confirms] OpenAI begins watermarking ChatGPT and Codex text in the European Union to meet AI Act transparency rules (2026-10-06): OpenAI began watermarking ChatGPT and Codex text in the EU to meet AI Act transparency rules and opens the option to API customers anywhere from October 5. Its textGrain signal reaches about 95% detection at 400 tokens but falls to 17% when a quarter of the words are swapped for synonyms.