加州SB 942生效:AI生成內容今起強制標示來源 | California SB 942 Live: AI Content Must Disclose Origin
By Kit 小克 | AI Tool Observer | 2026-08-03
🇹🇼 加州SB 942生效:AI生成內容今起強制標示來源
加州SB 942(California AI Transparency Act)今天(2026年8月2日)正式生效,這是美國第一個要求主流AI公司在生成內容中嵌入「來源標示」的州級法律。只要是月活躍用戶超過100萬的生成式AI服務,從今天起都必須讓使用者知道眼前的圖片、影片或語音是不是AI做的。
什麼是SB 942?誰需要遵守?
SB 942要求大型生成式AI服務商在輸出內容中嵌入可偵測的「來源資料」(provenance data),並提供免費的公開偵測工具讓一般人查驗內容真假。適用對象除了ChatGPT、Gemini這類大型AI工具,經修正案AB 853擴大後,還涵蓋月活躍用戶超過200萬的社群平台、提供模型權重下載的開源託管平台,以及內建攝影機、麥克風的硬體裝置製造商。
具體要揭露什麼?
- 可見標示(manifest disclosure):使用者介面上要有清楚的AI生成標記,可開可關
- 隱藏浮水印(latent disclosure):嵌入內容檔案裡、不易被肉眼發現的技術標記,符合C2PA規格
- 禁止移除工具:法律明文禁止製作或散布「去浮水印」的工具
- 免費偵測服務:公司要提供任何人都能用的內容溯源查驗工具
SB 942跟歐盟AI法案有什麼不同?
時間點不是巧合——AB 853特地把SB 942的生效日訂在今天,跟歐盟AI法案第50條(聊天機器人、Deepfake標示規定)同步上路。差異在於歐盟管的是「跟使用者互動時要說自己是AI」,加州SB 942管的是「輸出的內容本身要能被技術性驗證來源」,等於是從行為揭露走向內容鑑識,對Meta、Google、OpenAI這類同時在加州與歐盟都有龐大用戶的公司來說,等於同一天要應付兩套不同邏輯的合規要求。
對一般使用者有什麼實際影響?
短期內不會馬上感受到差異,因為大部分公司選擇低調合規而非大張旗鼓宣傳。但接下來幾個月,你在用Gemini、ChatGPT等工具產生圖片影片時,應該會開始看到內建的「AI生成」標記選項,社群平台也會陸續上線「查看內容來源」的功能。對內容創作者跟自媒體來說,這代表未來要更小心處理AI素材,避免踩到「移除標示」的紅線。
好不好用,試了才知道。
🇺🇸 California SB 942 Live: AI Content Must Disclose Origin
California SB 942 (the California AI Transparency Act) took effect today, August 2, 2026 the first US state law requiring major AI companies to embed detectable origin markers in generated content. Any generative AI service with more than 1 million monthly users must now let people check whether an image, video, or audio clip was made by AI.
What Does SB 942 Actually Require?
SB 942 requires large generative AI providers to embed detectable provenance data in their outputs and offer a free public tool anyone can use to check content authenticity. Its amendment, AB 853, expanded coverage beyond chatbots like ChatGPT and Gemini to include large social platforms with 2M+ monthly users, generative AI hosting platforms that distribute open-weight models, and hardware makers building devices with built-in cameras or microphones.
What Has to Be Disclosed?
- Manifest disclosure: a visible, toggleable AI-generated label in the user interface
- Latent disclosure: a technical watermark embedded in the file itself, compatible with C2PA standards
- Anti-stripping clause: the law explicitly bans building or distributing tools that remove these watermarks
- Free detection tool: providers must offer a public tool for anyone to verify content origin
How Does SB 942 Compare to the EU AI Act?
The timing is deliberate AB 853 moved SB 942's operative date to align with Article 50 of the EU AI Act, which also mandates chatbot and deepfake disclosure. The key difference: the EU rule focuses on behavioral disclosure (telling users they're talking to an AI), while California's SB 942 focuses on content forensics (making the output itself technically verifiable). For companies like Meta, Google, and OpenAI operating in both jurisdictions, today means satisfying two distinct compliance frameworks simultaneously.
What Changes for Everyday Users?
Nothing dramatic yet most providers are complying quietly rather than announcing it loudly. But over the coming months, expect to see built-in AI-generated labeling options appear when you create images or video in Gemini or ChatGPT, and social platforms rolling out content-origin-check features. For content creators, it means being more careful with AI-generated assets, since stripping provenance markers is now explicitly illegal in California.
好不好用,試了才知道 good or not, you won't know until you try it.
Sources / 資料來源
- SB-942 California AI Transparency Act — Full Bill Text
- Key Deadlines Under the California AI Transparency Act
- California AI Transparency Act (SB 942): 2026 Compliance Guide
常見問題 FAQ
SB 942是什麼時候生效的?
SB 942於2026年8月2日正式生效,修正案AB 853特地將日期對齊歐盟AI法案第50條的執行日。
SB 942適用哪些公司?
月活躍用戶超過100萬的生成式AI服務商,以及月活躍用戶超過200萬的大型社群平台、開源模型託管平台、內建攝影機或麥克風的硬體裝置製造商。
SB 942跟歐盟AI法案第50條差在哪?
歐盟規定AI要主動揭露自己是AI(行為揭露),加州SB 942則要求生成內容本身要能被技術性驗證來源(內容鑑識),兩者互補但方法不同。
一般用戶會看到什麼改變?
短期內不會有明顯變化,但幾個月內主流AI工具會陸續加入可見的AI標示選項,社群平台也會上線內容來源查驗功能。
延伸閱讀 / Related Articles
- DOGE AI工具SweetREX曝光:美國政府砍法規爭議大 | DOGE AI Tool SweetREX Exposed: US Deregulation Bot
- Kimi K3開源模型評測:2.8兆參數寫程式贏過Claude | Kimi K3 Review: China's 2.8T Model Beats Claude at Coding
- Claude Sonnet 5漲價倒數:帳單恐暴增50% | Claude Sonnet 5 Price Deadline: Bills May Jump 50%
AI 工具觀察站 — 每日精選 AI Agent 與工具趨勢
AI Tool Observer — Daily curated AI Agent & tool trends
留言
張貼留言