aifollow.news
SpaceXAI 将为 Grok 生成文本添加隐形水印
Automatically verified and published · Generated and evidence-checked automatically; not reviewed by a human.
SpaceXAI will begin adding invisible watermarks to Grok-generated text on November 7, allowing a detector to identify watermarked text. The report says watermarking is already enabled across SpaceXAI APIs worldwide for several Grok models, though team administrators can disable it for individual models. The watermark is built into token selection during text generation, rather than hidden characters or metadata.
IT之家 11 月 1 日消息，SpaceXAI 将于 11 月 7 日起为 Grok 模型生成的文本添加隐形水印。检测器通过识别该信号，判断一段文本是否带有肉眼不可见水印。
官方表示，本次调整是为了响应欧盟《人工智能法案》要求，全球所有地区的 SpaceXAI API 均已启用水印，包含 grok-4.7、grok-4.6、grok-4.5、grok-4.3、grok-4.20 reasoning、grok-4.20 non-reasoning 和 grok-4.20 multi-agent 模型，团队管理员可选择为某款模型单独关闭水印。
IT之家注意到，这种隐藏式水印并非通过隐藏字符或元数据实现，而是融入模型生成文本时对 Token 的选择过程。
AI 治理观察者安德鲁 · 库兰（IT之家注：Andrew Curran）则对此表示，目前还没有为文字加入隐形水印的模型仅有北美地区 ChatGPT。
Name a person, company or topic. We find the sources and gather every new development in your own tab.