OpenAI Implements Invisible Watermarking for ChatGPT in the EU
OpenAI will begin watermarking text generated by its AI models like ChatGPT and Codex within the European Union to comply with the EU AI Act. The regulation, effective since August 2, mandates that AI-generated content be marked so other systems can identify it. OpenAI's approach uses an invisible technique called 'textGrain,' which subtly influences word choices in a way detectable by specialized tools but not visible to readers. The watermark will roll out over the coming weeks for EU users across all plans, while API developers worldwide can enable it selectively starting now. The company emphasized that the watermark does not identify individual users and has observed no significant impact on model performance. However, OpenAI noted limitations: short texts or heavily edited content may be harder to detect. This move follows similar actions by Anthropic with its Claude AI system in August 2024 but differs as it is limited initially to EU users rather than a global rollout.
Topics: artificial intelligence EU AI Act regulation watermarking