由 2026 年 8 月 2 日起推出嘅 Claude 新模型,會喺生成文字入面加入肉眼睇唔到嘅統計式水印;支援嘅檔案格式就會加入數碼簽署嘅來源資料。
Anthropic 表示,水印唔包含任何可以識別個人、機構或者對話嘅資料。
移除工具好快引起關注:其中一個 GitHub 專案三日內吸引約 8,983 粒 stars,但呢個數字只反映人氣,唔代表已證實可以成功破解 Claude 未公開嘅統計式水印。
What happened after Anthropic began embedding a statistical watermark in all Claude-generated text and files worldwide on August 2, 2026 toAnthropic’s Claude watermark is designed as a machine-detectable provenance signal, not a conclusive authorship record.
AI 提示
Create a landscape editorial hero image for this Studio Global article: What happened after Anthropic began embedding a statistical watermark in all Claude-generated text and files worldwide on August 2, 2026 to. Article summary: The rollout triggered an immediate, mostly symbolic arms race: Anthropic’s mark is designed as a probabilistic signal of Claude involvement, while developers rapidly published tools claiming to erase it. But there is no . Topic tags: general, general web, user generated, government, academic. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, water
openai.com
Claude 新增嘅隱形水印,最準確嘅理解係一條內容來源線索,而唔係一個可以證明「邊個寫」嘅數碼簽名。
由 2026 年 8 月 2 日起推出嘅 Claude 模型,會喺生成文字時加入肉眼無法察覺嘅統計模式;至於部分支援嘅檔案,就會附帶數碼簽署嘅來源資料。Anthropic 因應歐盟《人工智能法案》(EU AI Act)嘅透明度要求,將安排套用到全球,而唔係只限歐洲用戶。
消息公布後,開源及瀏覽器工具幾乎即刻湧現,聲稱可以清走 AI 痕跡。不過最關鍵、亦最容易畀人忽略嘅一點係:Anthropic 暫時未公開完整技術規格,亦未提供一個公眾可以獨立測試嘅官方偵測器。因此,外界未有足夠條件驗證某個工具係咪真係成功擊破 Claude 水印。
由 2026 年 8 月 2 日起推出嘅 Claude 新模型,會喺生成文字入面加入肉眼睇唔到嘅統計式水印;支援嘅檔案格式就會加入數碼簽署嘅來源資料。
Anthropic 表示,水印唔包含任何可以識別個人、機構或者對話嘅資料。
移除工具好快引起關注:其中一個 GitHub 專案三日內吸引約 8,983 粒 stars,但呢個數字只反映人氣,唔代表已證實可以成功破解 Claude 未公開嘅統計式水印。
What happened after Anthropic began embedding a statistical watermark in all Claude-generated text and files worldwide on August 2, 2026 toAnthropic’s Claude watermark is designed as a machine-detectable provenance signal, not a conclusive authorship record.
AI 提示
Create a landscape editorial hero image for this Studio Global article: What happened after Anthropic began embedding a statistical watermark in all Claude-generated text and files worldwide on August 2, 2026 to. Article summary: The rollout triggered an immediate, mostly symbolic arms race: Anthropic’s mark is designed as a probabilistic signal of Claude involvement, while developers rapidly published tools claiming to erase it. But there is no . Topic tags: general, general web, user generated, government, academic. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, water
openai.com
Claude 新增嘅隱形水印,最準確嘅理解係一條內容來源線索,而唔係一個可以證明「邊個寫」嘅數碼簽名。
由 2026 年 8 月 2 日起推出嘅 Claude 模型,會喺生成文字時加入肉眼無法察覺嘅統計模式;至於部分支援嘅檔案,就會附帶數碼簽署嘅來源資料。Anthropic 因應歐盟《人工智能法案》(EU AI Act)嘅透明度要求,將安排套用到全球,而唔係只限歐洲用戶。
消息公布後,開源及瀏覽器工具幾乎即刻湧現,聲稱可以清走 AI 痕跡。不過最關鍵、亦最容易畀人忽略嘅一點係:Anthropic 暫時未公開完整技術規格,亦未提供一個公眾可以獨立測試嘅官方偵測器。因此,外界未有足夠條件驗證某個工具係咪真係成功擊破 Claude 水印。