OpenAI Forced to Watermark ChatGPT Text in EU Under Landmark AI Law

OpenAI announced on Monday that it will implement watermarking for text generated by its ChatGPT and Codex models within the European Union, a move aimed at complying with the EU's Artificial Intelligence Act. The watermarking, which embeds invisible markers in AI-generated text, is designed to help identify content produced by the company's AI systems.

The EU AI Act, which entered into force in August 2024, imposes transparency obligations on providers of general-purpose AI systems, requiring them to disclose AI-generated content. OpenAI's decision follows similar commitments from other major AI developers. However, the company cautioned that editing the text can make the invisible watermarks harder to detect, potentially undermining their reliability as a provenance tool.

The effectiveness of watermarking as a compliance mechanism remains debated among technologists and policymakers. While the EU has championed the AI Act as a global regulatory benchmark, critics argue that watermarking can be easily circumvented. OpenAI's move may pressure other AI firms to adopt similar measures, but it also highlights the challenges of enforcing transparency in rapidly evolving AI systems. The company has not yet announced a timeline for implementation or whether the watermarking will extend beyond the EU.