OpenAI Introduces Text Watermarking and Detection Tools in Response to EU AI Act

2026年10月5日，OpenAI公布了其合规策略，以满足欧盟《人工智能法案》中关于机器可读文本溯源的要求。该公司发布了一项名为textGrain的专有水印技术，可在生成的文本中嵌入肉眼不可见的统计信号。未来数周内，OpenAI计划面向欧盟地区用户，在符合条件的ChatGPT及Codex输出内容中部署此类隐形水印。全球API客户即日起可针对特定模型主动开启水印功能；与此同时，为评估检测可靠性并降低误报率，水印检测工具在初期将仅向获批准的研究人员和专业机构开放。

On October 5, 2026, OpenAI announced its strategy for complying with machine-readable text provenance requirements under the European Union AI Act. The company unveiled textGrain, a proprietary watermarking technology that embeds invisible statistical signals into generated text. In the coming weeks, OpenAI plans to deploy invisible watermarks across eligible ChatGPT and Codex outputs for users in the EU. Global API customers can immediately opt in to watermarking for select models, while watermark detection tools will initially be restricted to approved researchers and expert organizations to evaluate reliability and mitigate false positives.