OpenAI to watermark ChatGPT outputs by default in the EU under AI Act
Ars Technica AI · rss · 2026-10-07
OpenAI will automatically watermark ChatGPT-generated text in the EU, with the feature opt-in elsewhere, to comply with the EU AI Act that took effect in August and requires machine-detectable marking of AI content.
- The proprietary technique is called textGrain and is described in a technical paper. Like other LLM watermarking schemes, it embeds subtle patterns in word choices that don't degrade output quality but can be detected by a specialized detector with the right key.
- Existing standards like SynthID and C2PA are relatively easy to circumvent, and textGrain is likely similar.
- Detector access will go to a limited set of researchers and organizations, with an approval process for others over time.
More from Models
- Many Mistral Large 4 failures traced to reasoning mode not being enabled — qtnx_ · 2026-10-07
- Early Opus 5.5 user says hype is overblown: shortcuts, wrong assumptions, sloppy work — haider1 · 2026-10-07
- TypeSafe's Jev model bets on machine-native intelligence over text-optimized LLMs — TWIML AI Podcast · 2026-10-07
- llm-mistral 0.16 adds reasoning model support for Mistral Large 4 — Simon Willison · 2026-10-07
- Ollama hosts Google's EmbeddingGemma 2, a 740M multimodal embedding model for on-device use — ollama · 2026-10-07
- Reddit users grow frustrated with ChatGPT's over-refusals on innocuous prompts — Crixusgannicus · 2026-10-07