OpenAI将在欧盟给ChatGPT文本加隐形水印，应对AI法案
OpenAI 宣布，将在欧盟区对 ChatGPT 和 Codex 生成的文本、代码嵌入隐形统计水印，以满足欧盟《AI法案》12月2日的透明度大限。水印系统名为 textGrain。
OpenAI 计划开源水印实现方案，但检测软件初期仅限获批研究人员和专家机构使用——短文本或被大幅改写后的文本检测准确率会下降。
与此同时，苹果确认 macOS 将收紧“完全磁盘访问”权限：AI 助手类应用须经用户“非常明确的操作”才能获得全盘访问。此前有报道称 Meta 的 Muse 助手在未获明确许可的情况下读取了用户私密消息，引发隐私担忧。（综合 Engadget、PCMag 报道）
OpenAI to Watermark ChatGPT Text in EU Ahead of AI Act Deadline
OpenAI will embed an invisible statistical watermark — via its textGrain system — into ChatGPT and Codex text and code outputs across the EU to meet the EU AI Act's Dec. 2 transparency deadline.
OpenAI plans to open-source the watermarking implementation, though detection software will initially be limited to approved researchers and expert organizations, as accuracy drops on short or heavily edited text.
Separately, Apple confirmed macOS will require "very explicit user action" before apps get Full Disk Access, amid concerns over autonomous AI agents. The move follows reports that Meta's Muse agent accessed private messages without clear permission. (Via Engadget, PCMag)
图片来源：Wikimedia Commons