OpenAI to Watermark ChatGPT Text in EU

OpenAI plans to watermark ChatGPT and Codex text in the European Union as part of its efforts to comply with the EU AI Act. The watermark will be invisible and machine-readable, designed to help identify text generated by OpenAI's systems. According to OpenAI, the approach uses textGrain technology, which the company claims matched or exceeded other watermarking methods, including Google DeepMind's SynthID for text. The move comes amid growing regulatory pressure on AI transparency and content provenance. European rules are pushing AI providers to make it easier to tell when content has been generated or altered by AI. Watermarking is one way to address that requirement, though it is not a complete solution. OpenAI says detection access will initially be limited to researchers, which suggests the company is treating the capability carefully while it studies accuracy and misuse risks. There are practical limitations. OpenAI notes that editing can make watermarks harder to detect, meaning the system may be less reliable after users revise, paraphrase, or transform generated text. Watermarks also need to work across languages, formats, and writing styles without degrading output quality. If detection is too easy to remove, the mark offers limited value; if it is too robust, it may raise concerns about privacy or unintended tracking. Even with those caveats, the EU rollout could set a precedent for how generative AI companies handle provenance elsewhere. As AI-generated text becomes more common in journalism, education, customer service, and software development, regulators and platforms want better tools for identifying its origin. OpenAI's decision to watermark ChatGPT and Codex output in one major jurisdiction may encourage similar measures in other markets or become a baseline expectation for responsible AI deployment. The announcement also reflects a broader shift: transparency features are moving from research demos to compliance infrastructure. For OpenAI, watermarking is not just a technical feature but a regulatory strategy. For users and businesses, it signals that AI provenance may soon be part of everyday workflows, especially in regions with strict AI laws. How well textGrain performs in real-world conditions will determine whether watermarking becomes a trusted standard or simply one more layer in a complex content authenticity debate.