歐洲非營利組織 AI Forensics 發布報告指出，開源人工智慧平台 Hugging Face 存在嚴重的非自願性深偽問題。研究人員測試了該平台上九個熱門圖片編輯空間，其中七個能輕易將穿著衣物的女性圖片轉換為裸露圖片，僅需使用簡單的六個字提示詞即可完成，無需繞過任何安全機制。

研究人員進一步在 Hugging Face 上建立了蜜罐式圖片編輯空間，一週內收集超過一千條提示詞與圖片。結果顯示，73% 的提示詞具有色情性質，其中 83% 意圖脫衣或性化照片中的對象，95% 的目標為女性，更有 6.7% 的色情請求針對疑似未成年人。這些數據表明，使用者確實正在利用該平台生成非自願的親密影像。

面對指控，Hugging Face 表示研究方法可能導致誤判，並稱所蒐集的提示詞不具平台代表性，且因技術限制無法對所有空間進行提示詞過濾或輸出掃描。然而研究人員反駁稱，平台在明知缺乏審核能力的情況下部署這些空間，不應免除其對可預見後果的責任。與此同時，美國、歐盟和英國正加強打擊有害性深偽內容，包括查封相關網站並計劃禁止「裸化」應用程式。

A report by European nonprofit AI Forensics reveals that Hugging Face, a major open-source AI platform, has a pervasive problem with nonconsensual sexual deepfakes. Researchers tested nine popular image editing Spaces on the platform and found that seven could easily transform a clothed image of a woman into a topless one using a simple six-word prompt, with no safety guardrails in place at the platform level.

The researchers also deployed honeypot-style Spaces on Hugging Face and collected over 1,000 prompts in one week. Seventy-three percent of prompts were sexual in nature, with 83 percent seeking to undress or sexualize the submitted person—95 percent of whom were women. Alarmingly, 6.7 percent of sexual requests targeted apparent children. The prompts revealed abuse extending beyond simple digital undressing, including requests depicting sexual acts and the removal of hijabs from Muslim women.

Hugging Face questioned the research methodology, claimed the prompts were not representative of typical platform usage, and stated that platform-wide prompt filtering is technically infeasible. Researchers countered that the company should not be exempt from responsibility for deploying Spaces it cannot moderate. Meanwhile, US law enforcement has seized deepfake hosting sites, and the EU and UK are advancing plans to ban nudifying apps, signaling a broader regulatory crackdown on harmful sexual deepfake technologies.