To comply with the EU AI Act, OpenAI plans to add text watermarking for ChatGPT and Codex users in the EU and an opt-in setting for API customers globally

Generated text from OpenAI tools becomes identifiable as AI-produced in the EU, and globally optional for API users.

OpenAI plans to introduce text watermarking for its ChatGPT and Codex services to comply with the European Union's AI Act. This measure will involve embedding a hidden signal within the generated text, making it identifiable as AI-produced. The company will also offer an opt-in setting for API customers worldwide to enable this watermarking feature. The AI Act aims to increase transparency and accountability in AI systems, and this watermarking is a step towards meeting those regulatory requirements within the EU. The rollout is expected to begin for ChatGPT and Codex users in the EU first, with global API customers having the option to enable it.

This move by OpenAI signals a proactive approach to regulatory compliance, particularly concerning the EU AI Act's transparency requirements for generative AI. By implementing watermarking, OpenAI aims to provide a verifiable method for distinguishing AI-generated content from human-created text, which could impact content moderation, copyright discussions, and the spread of misinformation. The opt-in for global API customers suggests a tiered strategy, balancing EU legal obligations with broader market considerations and developer preferences.

The introduction of watermarking will likely influence how other AI developers approach compliance and content provenance, potentially setting a de facto industry standard. Users and organizations will need to adapt to identifying and managing AI-generated content, which could lead to new workflows and tools for verification. The long-term implications may include a more regulated landscape for AI-generated content and increased scrutiny on the accountability of AI providers.