←── back to feed
/topics/openai-text-watermarking-chatgpt-eu
OpenAI text watermarking ChatGPT EU
3 items●1 sources●updated 33d ago●trend 0
OpenAI is rolling out invisible, machine-readable text watermarks to ChatGPT and Codex outputs for EU users to comply with the AI Act's provenance requirements. The watermarking system, called textGrain, matches or exceeds competitors like Google DeepMind's SynthID and Anthropic's approach, though editing can degrade detection.
- textGrain watermark is invisible and machine-readable, embedded in ChatGPT and Codex text output
- Rollout limited to EU users initially to meet AI Act compliance requirements
- OpenAI claims textGrain performance matches or exceeds Google DeepMind's SynthID and Anthropic's August watermarking
- Editing watermarked text can make invisible marks harder to detect, OpenAI acknowledges
- Benchmark scores show watermarked and unwatermarked text perform similarly