OpenAI to Implement ChatGPT Watermarking in EU, Optional for Global API Users

OpenAI will implement watermarking for ChatGPT text in the EU starting in the coming weeks, while allowing API users globally to opt-out. This watermarking technology, developed in collaboration with Microsoft Azure, aims to enhance content traceability without degrading output quality, as claimed by OpenAI. Detection rates for watermarked text vary significantly based on text length and subject matter, achieving a 95% detection rate at 400 tokens in psychology-related content, but dropping to below 20% with 25% word replacements. OpenAI's Codex and Astra models are also expected to incorporate similar watermarking features. This decision follows OpenAI's ongoing commitment to ethical AI development, as highlighted in recent announcements. Competing AI provider Anthropic, known for its Claude model, has not yet announced similar watermarking initiatives. The activation of watermarking for EU users aligns with regulatory pressures, particularly from the EU AI Act, which emphasizes transparency in AI-generated content.