OpenAI is set to embed invisible watermarks in all ChatGPT and Codex outputs distributed in the European Union, a move aimed at satisfying the new AI Act’s transparency requirements.
Watermarking in the EU: Key Details
- All ChatGPT and Codex text will carry an invisible digital watermark.
- The marks are designed to trace the source of AI‑generated content.
- Users can edit text, potentially making the watermark harder to detect.
- OpenAI says the system will still function under the AI Act’s disclosure rules.
Why Watermarking Matters: A Regulatory Backdrop
The EU’s AI Act, which came into force in 2024, mandates that high‑risk AI systems disclose their use and provide traceability. OpenAI’s move follows similar steps by other firms, such as Anthropic and Google Gemini, to embed provenance markers. Watermarking has been tested in academic settings and is now being rolled out at scale to meet legal obligations and bolster user trust.
What Comes Next: Industry and User Implications
While the watermarking feature enhances accountability, it also raises questions about content editing and the robustness of detection tools. Regulators will likely monitor how effectively the marks survive transformations, and developers may need to refine algorithms to maintain traceability. The rollout could set a precedent for other AI providers, shaping the future of compliance and transparency in the EU market.