主題樞紐TOPIC HUB
AI 監管與政策AI Policy & Regulation
全站 AI 監管與政策文章總覽:各國立法與執法動向、資料與隱私規則、主權與路由要求,以及合規對產品團隊的實際影響。Every post here on AI policy: legislation and enforcement moves, data and privacy rules, sovereignty and routing requirements, and what compliance actually means for product teams.
主題簡介ABOUT THIS TOPIC
158 篇文章158 ARTICLES這個主題收集全站與 AI 監管、政策、合規相關的文章:美國州級與聯邦法案、歐盟 DSA/AI Act 的執行進度,以及隱私、資料保留與主權路由等規則對 AI 產品的實際約束。
除了立法進度,這裡也收錄治理面向的分析:模型不當行為的通報框架、受監管產業的採購前提,以及監管與創新之間的取捨。
必讀精選優先收錄對產品與工程決策有直接影響的政策分析;時間線收錄全部相關文章,新文在前。
This hub collects the coverage of AI regulation, policy, and compliance: US state and federal bills, the enforcement pace of the EU DSA and AI Act, and the practical constraints that privacy, data-retention, and sovereignty-routing rules place on AI products.
Beyond legislative tracking, it includes governance analysis: misalignment reporting frameworks, procurement preconditions in regulated industries, and the trade-offs between oversight and innovation.
The must-read picks favor policy analysis with direct product and engineering consequences; the timeline lists every related post, newest first.
必讀精選MUST READ
6- Anthropic 與 OpenAI 的州級 AI 安全法案之戰:棘輪式升級 vs 反向聯邦主義美國國會立法停擺下,AI 實驗室轉戰州議會。Anthropic 力推各州逐級加嚴(伊利諾州已簽署全美首見的第三方稽核法),OpenAI 則主打「同一套法案複製到各州」;兩家在伊利諾州 SB 3444 免責條款上正面交鋒。本文整理雙方策略與各州進展。閱讀 ↗
- 歐盟將 ChatGPT 列為 VLOSE:首個進入 DSA 最嚴格層級的 AI 聊天機器人2026 年 8 月 31 日,歐盟執委會依數位服務法將 ChatGPT 指定為超大線上搜尋引擎(VLOSE),成為首個進入 DSA 最嚴格層級的 AI 聊天機器人;Reddit 與 Roblox 同日被列為 VLOP。閱讀 ↗
- 模型不聽話時,該怎麼記錄、調查、公開?OpenAI 的 misalignment 通報框架OpenAI 提出一套模型偏差通報框架,從發現、調查到揭露,讓開發者能更快分享異常行為,即使還沒找到原因或解法。閱讀 ↗
- Claude 免費用戶的資料開關:訓練授權、五年保留期與產品團隊要留意的界線Anthropic 更新消費者條款,讓 Claude Free、Pro、Max 用戶自行決定是否用對話資料訓練模型,並把保留期從 30 天延長到五年。閱讀 ↗
- 把 agentic AI 帶進政府與受監管產業:Cohere 與 OpenText 的合作對產品團隊的意義Cohere 與 OpenText 合作,讓 agentic AI 能在本地、私有雲或主權雲部署,直接影響產品團隊的架構選擇與採購決策。閱讀 ↗
- 加州年齡驗證法修法豁免開源系統:AB 1856 解讀加州參議院以 39 票對 0 票通過 AB 1856 修正案,把 GPL、MIT、BSD、Apache 授權的作業系統排除在年齡驗證義務外,全案已送交州長紐森簽署。本文解讀條文與影響。閱讀 ↗
- Anthropic vs OpenAI on State AI Safety Bills: The Ratchet Against Reverse FederalismUS AI rules moved to the states: Anthropic pushes ever-stricter bills while OpenAI replicates one template, colliding over Illinois SB 3444's liability shield and the SB 3261 audit mandate.READ ↗
- EU Designates ChatGPT a VLOSE: The First AI Chatbot in the DSA's Strictest TierOn August 31, 2026, the EU Commission designated ChatGPT a Very Large Online Search Engine under the DSA, the first AI chatbot in the strictest tier; Reddit and Roblox became VLOPs the same day.READ ↗
- A Misalignment Disclosure Process You Can Actually CopyOpenAI's misalignment reporting framework sets disclosure criteria, tracks, and report fields builders can adapt.READ ↗
- What Anthropic's Consumer Data Opt-In Changes for How You Build on ClaudeAnthropic now lets Claude Free, Pro, and Max users opt into training data use, with a 5-year retention window and an Oct 8 deadline.READ ↗
- What Cohere and OpenText's Partnership Changes for Shipping Agentic AI in Regulated SettingsA concrete look at how combining OpenText data context with Cohere's North platform shifts deployment options for builders in regulated industries.READ ↗
- California Exempts Linux from Age-Verification LawCalifornia passed AB 1856 unanimously, exempting operating systems under GPL, MIT, BSD, and Apache licenses from age-verification duties. The bill now awaits Governor Newsom.READ ↗
完整時間線
158 篇文章展開全部 158 篇文章
2026 / 143 篇
- 當推論必須進入受信任環境:NVIDIA 機密運算對推論架構的實際影響NVIDIA 機密運算讓 LLM 推論跑在記憶體加密的機密 VM 與機密 GPU 上,產品團隊要重新盤算效能與隱私的取捨。繁中
- 從所有權結構下手:Anthropic 收緊不支援地區銷售限制對產品團隊的意義Anthropic 把限制從「在哪裡營運」改成「誰控制你」,產品團隊做企業採購與合規時要重新檢視客戶的所有權結構。繁中
- 當安全標準變成入場門檻:Cohere 對 AI 規則制定權的質疑Cohere 執行長 Aidan Gomez 質疑少數實驗室以安全之名取得反壟斷豁免,並提出四項替代原則。繁中
- 青少年安全藍圖不是合規清單:產品團隊要提早設計的六個決策OpenAI 發布澳洲青少年安全藍圖,把年齡保證、家長控制與危機轉介寫成產品預設,而不只是政策文件。繁中
- 把 agentic AI 帶進政府與受監管產業:Cohere 與 OpenText 的合作對產品團隊的意義Cohere 與 OpenText 合作,讓 agentic AI 能在本地、私有雲或主權雲部署,直接影響產品團隊的架構選擇與採購決策。繁中
- 當生物研究被模型擋下:Anthropic 生命科學驗證計畫想換掉什麼Anthropic 推出生命科學驗證計畫,以機構審核換取較寬鬆的生物類模型存取,並把即時攔阻改成離線監控。繁中
- 模型不聽話時,該怎麼記錄、調查、公開?OpenAI 的 misalignment 通報框架OpenAI 提出一套模型偏差通報框架,從發現、調查到揭露,讓開發者能更快分享異常行為,即使還沒找到原因或解法。繁中
- Claude 免費用戶的資料開關:訓練授權、五年保留期與產品團隊要留意的界線Anthropic 更新消費者條款,讓 Claude Free、Pro、Max 用戶自行決定是否用對話資料訓練模型,並把保留期從 30 天延長到五年。繁中
- 當瀏覽器內建 AI:Mistral 與 Mozilla 把主權與隱私放進 Firefox Smart WindowMistral 與 Mozilla 合作,讓 Firefox Smart Window 由 Mistral 模型驅動,主打零資料保留與在地語言微調。繁中
- 當 AI 走進國防與公部門:Anthropic 顧問團對產品團隊的三個現實提醒Anthropic 成立國家安全與公部門顧問團,本文拆解這對做 AI 產品的人在合規、部署與標準上的實際影響。繁中
- ZDR 不是隱私萬靈丹:把「零保留」當成可強制執行的路由條件ZDR 只保證推論供應商不儲存 prompt 與回應,不涵蓋你的日誌、工具或快取;用 provider.zdr 把它變成請求層級的硬性路由條件。繁中
- 區域路由的關鍵不是「在哪推論」,而是「請求在哪解密」OpenRouter 推出 US/EU In-Region Routing,讓資料從解密到推論全程留在指定區域,避免合規審查翻車。繁中
- Oklo 告上 FERC:750 MW 資料中心電廠卡在 PJM 併網隊列核能新創 Oklo 向 FERC 提緊急申訴,要求 PJM 恢復其 750 MW 混合電廠(150 MW 先進核能、300 MW 燃料電池、300 MW 燃氣)的併網審查資格,案外人都在問:AI 資料中心搶電,卡在電網排隊。繁中
- Anthropic 與 OpenAI 的州級 AI 安全法案之戰:棘輪式升級 vs 反向聯邦主義美國國會立法停擺下,AI 實驗室轉戰州議會。Anthropic 力推各州逐級加嚴(伊利諾州已簽署全美首見的第三方稽核法),OpenAI 則主打「同一套法案複製到各州」;兩家在伊利諾州 SB 3444 免責條款上正面交鋒。本文整理雙方策略與各州進展。繁中
- GitHub Copilot 的九月雙重調整:帳務先付後用,體驗預設全開GitHub 於 2026 年 9 月 1 日重開 Business/Enterprise 信用卡新客並改為先付款後開通座位,10 月起既有客戶改期初預收;9 月 28 日起將 cloud agent、網頁與 Mobile Chat 合併為預設開啟的統一體驗,Chat 資料保留從 28 天延長為帳戶生命週期。管理員需要在 9/28 前做完三件事。繁中
- 加州年齡驗證法修法豁免開源系統:AB 1856 解讀加州參議院以 39 票對 0 票通過 AB 1856 修正案,把 GPL、MIT、BSD、Apache 授權的作業系統排除在年齡驗證義務外,全案已送交州長紐森簽署。本文解讀條文與影響。繁中
- 歐盟將 ChatGPT 列為 VLOSE:首個進入 DSA 最嚴格層級的 AI 聊天機器人2026 年 8 月 31 日,歐盟執委會依數位服務法將 ChatGPT 指定為超大線上搜尋引擎(VLOSE),成為首個進入 DSA 最嚴格層級的 AI 聊天機器人;Reddit 與 Roblox 同日被列為 VLOP。繁中
- Claude for Teachers:把教學標準和備課時間,還給老師Anthropic 推出 Claude for Teachers,為美國 K-12 教師提供免費的 Claude 高階功能、教學技能庫,並串接全美 50 州的學術標準與實證課程。本文整理產品重點、資料隱私設計,以及對教育科技建構者的啟示。繁中
- Anthropic 推出 2 億美元研究基金,探索 AI 經濟衝擊的應對方案Anthropic 宣布投入 2 億美元成立 Economic Futures Research Fund,支持外部研究機構進行大規模實驗與試點,以了解哪些政策與計劃能讓經濟在 AI 衝擊下更具韌性,並讓更多人分享 AI 帶來的利益。繁中
- Anthropic 撥款五百萬美元,資助 AI 對幸福感影響的獨立評估Anthropic 推出五百萬美元資助計劃,支持獨立研究 AI 對用戶幸福感的影響,並公開評估指引,涵蓋多輪對話、臨床專家參與、評分驗證、申請時程與成果開源要求。繁中
- Meta 兒少安全訴訟開庭:吹哨者作證,祖克柏早被告知Meta 前安全工程師 Arturo Béjar 在奧克蘭聯邦法庭作證,稱公司對兒少傷害採「不要問、不要說」態度,內部調查顯示 51% 青少年一週內遇過有害經驗;29 州檢察總長求償上看 2,000 億美元。繁中
- AI 如何改變權力平衡:OpenAI 新團隊的長期思考OpenAI 成立 Strategic Futures 團隊並推出 AI Futures 部落格,探討 AI 對自由社會權力結構的影響。文章從歷史、政治經濟學與制度設計角度,提出六項原則,並強調「權力平衡」而非「完全去中心化」才是關鍵。繁中
- 澳洲 50 萬戶家用電池:一年把批發電價砍了 47%澳洲家用電池補貼上路約一年突破 50 萬戶安裝,能源部長稱批發電價 12 個月下跌約 47%;在 AI 資料中心推高電力需求的年代,澳洲示範了分散式儲能壓低電價的路徑。繁中
- 伊利諾州把年齡驗證寫進作業系統:匿名上網再退一步伊利諾州 HB5511 於 7 月 31 日簽署:2028 年起作業系統須收集出生日期、以加密 API 向應用程式提供年齡區間,且沒有開源豁免;同週調查報導描繪了背後的遊說網絡。繁中
- 美國付 12.2 億美元請 RWE 放棄離岸風電租約德國能源巨頭 RWE 同意放棄紐約、加州與路易斯安那外海風電租約,換取美國內政部 12.2 億美元付款。這是川普政府第五筆同類交易,累計金額逼近 40 億美元。繁中
- Amazon 西德州 GW Ranch:7.65GW 燃氣電廠與全美最大碳排許可Amazon 在德州 Pecos County 買地申照,興建 AI 資料中心園區 GW Ranch,自建 7.65GW 燃氣電廠、初期不接德州電網;空污許可放行 3,300 萬噸二氧化碳,恐成全美最大單一污染源。繁中
- OpenAI 與司法部和解 320 萬美元:綠卡招聘歧視美國勞工案OpenAI 與子公司 Statsig 以 320 萬美元與美國司法部和解,案由是綠卡招聘程序中偏好臨時簽證持有者、排擠美國勞工,並接受招聘流程整改與監督。繁中
- 美國德州暫停資料中心併網審批:474GW 佇列迎來全面審計2026 年 8 月 3 日,德州州長 Abbott 指示 PUCT 與 ERCOT 對所有申請併網的資料中心展開全面查核,未通過者拒絕併網;背後是超過 474 GW、約九成來自資料中心的併網佇列。繁中
- 中國開始量產沉浸式 DUV 微影設備:年產 5 台的長征The Information 報導中國已開始量產沉浸式 DUV 微影設備,今年交付 SMIC、華虹與 CXMT,2026 年目標約 5 台、2027 年約 20 台;ASML 股價一度跌 8%,但分析師提醒良率與量產規模才是真正關卡。繁中
- Gemini Robotics 2 進場:從腳到指尖的機器人全身智能2026 年 7 月 30 日 Google DeepMind 發表 Gemini Robotics 2,以 VLA、ER 與 On-Device 三個模型讓人形機器人全身控制、綁繩結、組隊合作,並公布 ASIMOV-Agentic 安全基準。繁中
- OpenAI 在歐洲的負責任 AI 實踐:從框架到落地因應歐盟 AI 法下一階段,OpenAI 說明如何調整安全、透明與溯源機制:簽署 GPAI 與內容透明度行為準則、公開系統卡與 Model Spec,並以網路安全為例展示動態治理的實際做法。繁中
- Dario Amodei 表態:Anthropic 從未主張禁用開放權重模型美國官員傳出考慮禁用中國開放權重模型、科技業連署力挺開放權重之際,Amodei 於 7 月 27 日撰文澄清 Anthropic 從未主張禁令,改提出晶片出口管制、打擊工業級蒸餾與強制安全測試三項主張。繁中
- Google Play 年齡訊號 API 全球推廣:家長一處設定,App 自動調適內容Google Play 於 7 月 30 日宣布把 Age Signals API 從巴西推向全球:家長可在 Family Link 選擇分享子女年齡區間,開發者據此提供分齡體驗,8 月中先到澳洲與加拿大,年底前全面上線。繁中
- 歐盟登記「停止殺死網際網路」倡議:反對強制數位身分與年齡驗證歐盟委員會於 7 月 22 日登記歐洲公民倡議「Stop Killing The Internet」,要求數位身分與年齡驗證系統必須自願、保護隱私且不歧視,反對強迫公民使用特定數位錢包。本文解析訴求、程序與後續觀察重點。繁中
- 印度下令 GitHub 下架 Bitchat:藍牙通訊遇上審查印度內政部網路犯罪協調中心 I4C 於 2026 年 7 月 23 日發函,要求 GitHub 在三小時內停用 Jack Dorsey 的藍牙網狀通訊 app Bitchat 的三個儲存庫。通知書由 Dorsey 公開,事件凸顯審查權力伸進開源程式碼的新界線。繁中
- 機場一組密碼清空手機:GrapheneOS 用戶遭聯邦起訴亞特蘭大男子 Sam Tunick 在機場邊檢輸入 GrapheneOS 壓力碼清空手機,遭美國司法部以「為防止扣押而毀損財產」起訴,安全功能本身首次成為起訴焦點。繁中
- Anthropic 15 億美元圖書和解案獲批:每本書 3,000 美元聯邦法官 Araceli Martinez-Olguin 最終批准 Anthropic 15 億美元圖書侵權和解,約 50 萬本書、每本 3,000 美元;但 AI 訓練資料的版權之戰才剛進入下一輪。繁中
- 輝達、微軟、Meta 公開信:別過早設限開放權重模型輝達、微軟、Meta、Palantir 等逾 20 家公司連署公開信,反對對開放權重 AI 模型的過早限制;同一週白宮顧問指控 Kimi K3 蒸餾 Anthropic 技術,政策攻防升溫。繁中
- 近 200 家新創連署:別封殺中國開源權重模型Little Tech Association 在 7 月 22 日代表近 200 家公司致函川普政府,反對限制中國開源權重模型;新創擔心成本暴增、公司倒閉,OpenAI 與 Anthropic 則在華府聯手警告開源權重風險,兩條陣線正面交鋒。繁中
- 青少年與 AI:OpenAI 如何在安全與學習之間取平衡OpenAI 發布青少年 AI 使用方針:近九成青少年每週在 ChatGPT 上學習,官方主張「接取必須伴隨保護」。本文整理 Study Mode 的引導式學習設計、家長控制與高風險通知、休息提醒等防護機制,以及四項治理原則與外部合作。繁中
- 紐約州按下暫停鍵:50 MW 以上資料中心新案停審紐約州長 Hochul 簽署行政命令,成為全美第一個暫停大型資料中心審批的州:50 MW 以上的新許可全面停審,直到約一年後環評完成,電價與電網壓力是主因。繁中
- Claude 可能要看你的證件:Anthropic 驗證政策與生物特徵爭議TechCrunch 報導,Anthropic 更新隱私政策(7 月 8 日生效),特定情況下可要求 Claude 使用者上傳護照或駕照加自拍,並產生臉部幾何模板驗證身分,引發生物特徵資料保留與管轄權疑慮。繁中
- 荷蘭部長親赴華府反對 MATCH 法案:ASML 出口管制之戰升級荷蘭貿易部長 Sjoerdsma 赴華府遊說反對美國 MATCH 法案,該法將連 ASML 已獲准出售中國的 DUV 設備一併封鎖,而中國占 ASML 系統銷售約 19%,歐美在晶片出口管制上的矛盾正式浮上檯面。繁中
- FERC 令六大電網為 AI 資料中心開快車:30 天盤點、60 天交代費率FERC 委員一致通過命令,要求六大電網營運商加速資料中心等大負載併網:30 天內提報剩餘容量、60 天內辯護或修正費率,併網成本由資料中心自付。本文解析命令內容、電價壓力,與命令沒有解決的發電缺口。繁中
- 報導指出 SK Telecom 是 Claude Fable 5 出口管制事件的導火線2026 年 6 月 19 日,報導指出 SK Telecom 是美國 6 月暫停 Claude Fable 5 出口管制事件的真正觸發者。本文回顧 6 月 12 日對 Fable 5 與 Mythos 5 的管制與停權,並談對 model 供應風險的啟示。繁中
- 美國《Great American AI Act》草案:三年預佔州權引爆反彈美國眾議員 Obernolte 與 Trahan 發布 269 頁《Great American AI Act》討論草案:前沿模型須提安全計畫並接受半年一次第三方稽核,但三年州法預佔權引發勞工、消費者團體與民主黨內反彈。本文解析條文、時程與對開發者的意義。繁中
- Midjourney Medical 全身超音波掃描儀:60 秒成像的豪賭與爭議Midjourney 於 2026 年 6 月 18 日發表 Medical 全身超音波掃描儀:受測者浸入水槽,環形探頭 60 秒取滿全身資料,由 AI 重建成類 CT 影像。官方稱速度近百倍於 MRI,並引用「避免 30% 死亡」的說法,引發放射科醫師強烈質疑。繁中
- OpenAI 遭多州檢察長聯盟調查:傳票直指廣告、諂媚與未成年保護《華爾街日報》揭露美國多州檢察長組成聯盟調查 OpenAI,紐約州已送達傳票,範圍涵蓋廣告手法、使用者黏著、模型諂媚、健康資料與未成年人及長者保護,距離 OpenAI 秘密遞交 IPO 申請僅四天。繁中
- Anthropic 被要求停用 Fable 5 與 Mythos 5:一次關於監管與 AI 安全的警鐘美國政府以國家安全為由,要求 Anthropic 暫停所有用戶對 Fable 5 與 Mythos 5 的存取。Anthropic 公開回應,質疑此決定的技術依據,並指出此舉可能開創監管先例。本文整理事件始末,並探討對 AI 產品開發者的啟示。繁中
- 美國人點睇 AI?Anthropic 首份 Public Record 調查的 5 個關鍵發現Anthropic 發布首份 Public Record 調查,訪問近 52,000 名美國人。結果顯示:疾病治療是最大期望、失業是最普遍擔憂、跨黨派支持監管,但僅 15% 信任 AI 公司。本文整理對產品建構者的啟示。繁中
- xAI 遭前工程師提告:警示 Grok 安全風險反被逼退2026 年 6 月 10 日,前 xAI 工程師 Devin Kim 對 xAI 與 SpaceX 提告,稱因警示 Grok 的歧視與武器擴散風險遭報復逼退;他甫於 6 月初出任 Center for AI Safety 院長。繁中
- 五角大廈 1260H 清單增至 188 家:阿里、百度、比亞迪入列2026 年 6 月 8 日,五角大廈將 1260H「中國軍事企業」清單更新至 188 家,新增阿里巴巴、百度、比亞迪、蔚來與 Unitree 等公司,中國 AI 主力幾乎全數入列,美中科技脫鉤再升級。繁中
- Apple 拒在歐盟推出 Siri AI:DMA 豁免遭駁回始末Apple 宣布 Siri AI 不在歐盟推出,歸咎 DMA 要求「近乎無限的裝置存取」;執委會反擊「這是 Apple 自己的決定」,並證實 Apple 申請免除互操作性義務遭拒。iOS 27 今秋上路時,歐盟使用者將拿不到新版 Siri。繁中
- xAI 要求法院揭露 Grok 深偽原告身分:匿名訴訟權之戰Grok 深偽集體訴訟中,xAI 聲請揭開四位匿名原告的本名,主張圖片將封存、揭露無妨;原告律師斥「剝奪衣服後還要剝奪假名」,四位原告揚言被迫具名就退出。本文解析這場匿名權攻防對隱私訴訟的影響。繁中
- 英國 CMA 首度對 Google 下令:出版商可退出 AI 搜尋聚合2026 年 6 月 3 日,英國競爭及市場管理局對 Google 搜尋祭出首道行為要求:出版商可透過 Search Console 退出 AI Overviews 與 AI Mode、拒絕內容用於模型微調,並獲得清楚歸因。這是全球首見的 AI 搜尋退出權。繁中
- OpenAI 的政治立場:不設 PAC、不捐款,強調透明與直接倡議OpenAI 在 2026 年 6 月 1 日發布政策聲明,說明公司不設立政治行動委員會、不捐款給候選人或超級 PAC,並強調員工個人政治參與與公司立場分開。繁中
- 「This is fine」作者與 Artisan 達成協議:迷因挪用之爭落幕「This is fine」原作者 KC Green 指控 AI 新創 Artisan 在公車與地鐵廣告挪用其作品後,雙方於五月底達成協議:Artisan 下架紐約與舊金山廣告,Green 撤下抗議貼文。一起教科書級的 AI 行銷授權案例。繁中
- 中國傳要求頂尖 AI 人才出境先報准:阿里巴巴、DeepSeek 在列Bloomberg 報導,中國將出境限制擴及民營企業頂尖 AI 人才,阿里巴巴與 DeepSeek 員工出國前須取得官方核准。本文解析從 2025 年「建議」走向制度化的演變與對產業的影響。繁中
- Waymo 暫停亞特蘭大服務:機器計程車為何一再開進淹水路段2026 年 5 月 21 日,Waymo 因自駕車持續駛入淹水路段暫停亞特蘭大與聖安東尼奧服務,達拉斯與休士頓也預防性停駛。5 月中的軟體召回未解決問題,NHTSA 已介入了解。繁中
- FTC 重罰「主動傾聽」AI 廣告:從未監聽的 93 萬美元和解2026 年 5 月 21 日,FTC 宣布 Cox Media Group 等三家公司支付 93 萬美元和解:它們以「主動傾聽」為名販售 AI 廣告,宣稱能監聽對話投放廣告,實際上從未使用任何語音資料,只是加價轉售電子郵件名單。繁中
- arXiv 重拳整治 AI 垃圾論文:未查核內容者禁投一年2026 年 5 月中,arXiv 宣布對提交明顯未查核 AI 內容的作者處以一年禁令,解禁後投稿須先獲同儕審查場域接受。本文解析鐵證標準、處分程序與對預印本生態的影響。繁中
- Musk 訴 OpenAI 案倒數:Altman 出庭自辯「騙子」指控2026 年 5 月 12 日,Altman 在奧克蘭聯邦法院出庭作證,面對「偷走慈善機構」的主張與一長串「說謊者」名單的反詰問。Musk 求除去 Altman 與 Brockman 職務、重新分配 1,340 億美元。本文整理雙方證詞要點與結辯前的局勢。繁中
- 馬里蘭為外州 AI 資料中心多背 20 億美元電網帳單馬里蘭公用事業消費者代表辦公室向 FERC 投訴:PJM 三年 220 億美元輸電升級中,20 億美元落在馬里蘭,十年多付 16 億美元、住宅用戶約 345 美元,而負載成長幾乎都在外州資料中心。繁中
- Medicare ACCESS 給付新制:AI 醫療首次有了可收費的位置CMS 的 ACCESS 模型 7 月 5 日上線,150 個組織改以健康結果請領給付,AI 代理首次能在 Medicare 體系內自負盈虧。本文解析 Pair Team 的 Flora 語音代理、低給付率的設計邏輯,以及隱私與 CBO 超支的前車之鑑。繁中
- 賓州起訴 Character.AI:聊天機器人假冒精神科醫師2026 年 5 月 5 日,賓州 Shapiro 政府依《醫療執業法》起訴 Character.AI:臥底調查中聊天機器人自稱持照精神科醫師並出示捏造的執照號碼,州府尋求初步禁制令,是全美首件針對聊天機器人假冒醫療人員的訴訟。繁中
- 微軟據報考慮放棄 24/7 潔淨能源全時匹配目標:AI 用電壓力浮上檯面彭博報導微軟正評估延後或放棄 2030 年 24/7 潔淨能源全時匹配目標,AI 資料中心用電是關鍵壓力。本文解析該目標的技術意義、微軟官方回應,以及 hyperscaler 轉向天然氣對減排承諾的衝擊。繁中
- GPT-5.5 Instant 的系統卡透露了什麼:首次被列為高能力的 Instant 模型OpenAI 在 2026 年 5 月 5 日發布 GPT-5.5 Instant 系統卡,這是首個被列為高能力等級的 Instant 模型,特別在網路安全和生化防護方面。本文為產品開發者解析這份文件的重點與含義。繁中
- 哈佛急診研究:o1 診斷準確率勝過兩位主治醫師哈佛醫學院團隊在《Science》發表研究:在 76 名急診患者的檢傷診斷上,o1 以 67% 的精確或接近精確診斷率勝過兩位主治醫師的 55% 與 50%。本文解析數據、批評與落地限制。繁中
- NIST CAISI 評測 DeepSeek V4 Pro:距美國前沿約八個月美國商務部 NIST 旗下的 CAISI 發布 DeepSeek V4 Pro 評測:IRT 估計 Elo 800,約當八個月前的 GPT-5,數學幾乎追平、抽象推理與資安最弱,七項基準中五項比 GPT-5.4 mini 便宜。繁中
- 中國法院裁定:以 AI 取代員工不再是解僱的合法理由杭州中級人民法院判定,公司以 AI 接手工作後資遣員工屬違法解僱。這是六個月內第二起同類判決,確立「導入 AI 是商業選擇、非客觀情事變更」的見解,企業須改以再訓練或轉調因應。繁中
- 白宮擬 AI 監管行政命令:模型上市前審查成選項NYT 報導川普政府討論以行政命令成立 AI 工作小組,研究上市前模型審查;同週 CAISI 宣布 Google、Microsoft、xAI 加入預先評測。從撤除拜登規則到考慮審查,政策轉向值得開發者關注。繁中
- Google 開放五角大廈機密網路使用 AI:Anthropic 拒絕後的第三張門票2026 年 4 月 28 日,媒體報導 Google 同意五角大廈在機密網路使用其 AI,實質允許「一切合法用途」。本文解析合約的排除條款、Anthropic 因拒絕被列供應鏈風險的前車之鑑,以及 950 名 Google 員工的公開連署。繁中
- OpenAI 取得 FedRAMP Moderate 授權:美國政府採用 AI 的門檻降低OpenAI 的 ChatGPT Enterprise 與 API Platform 通過 FedRAMP 20x Moderate 授權,美國政府機構可更快速、安全地採用先進 AI。本文解析此里程碑的意義、FedRAMP 20x 流程的改變,以及對產品建構者的啟示。繁中
- 中國否決 Meta 併購 Manus:20 億美元交易踩線被撤中國發改委下令 Meta 與 AI Agent 新創 Manus 撤銷約 20 億美元的併購交易,是發改委首度以境內投資審查否決大型科技收購。本文整理半年審查時間軸、創辦人出境限制,與北京對 AI 資產出海劃下的新紅線。繁中
- Neurable 開放授權非侵入式 BCI:把腦波感測做成標準零組件2026 年 4 月,Neurable 宣布將非侵入式腦機介面技術授權給消費穿戴廠商:EEG 感測加上 AI 分析腦波評估認知表現,可嵌入耳機、帽子與頭帶。本文解析其授權模式、HyperX 合作與神經資料隱私。繁中
- Meta 員工鍵擊成訓練資料:Model Capability Initiative 的隱私爭議2026 年 4 月,Meta 向美國員工推出 Model Capability Initiative,記錄鍵擊、滑鼠軌跡與畫面內容來訓練電腦操作代理,CTO 明言無法退出。本文解析計畫內容、Muse Spark 模型與企業隱私課題。繁中
- Clarifai 刪除 300 萬張 OkCupid 照片:FTC 和解的真正代價路透報導,在 FTC 與 OkCupid 母公司 Match Group 和解後,AI 公司 Clarifai 刪除 2014 年取得的約 300 萬張使用者照片,以及用這批資料訓練的臉部辨識模型。本文解析七年調查始末,與訓練資料來源成為法律責任的訊號。繁中
- 美國 EIA 將強制資料中心申報用電:AI 電力足跡首度納入聯邦統計美國能源資訊局(EIA)向參議員證實,兩階段試點後將推出全國強制的資料中心用電調查:先對 196 家公司試點,蒐集年用電量、表後發電與冷卻系統等資料,AI 算力的電力足跡首度被聯邦量化。繁中
- 華府催華爾街測試 Anthropic Mythos:訴訟與合作並行的雙軌Bloomberg 揭露,財政部長 Bessent 與聯準會主席 Powell 召集銀行高層,促測 Anthropic 限流釋出的 Mythos 偵測資安漏洞;共同創辦人 Jack Clark 證實已向政府簡報,一邊控告國防部、一邊深度合作。繁中
- Novartis 執行長入董事會:Anthropic 信託任命董事過半Anthropic 宣布由長期利益信託任命 Novartis 執行長 Vas Narasimhan 為董事,讓信託任命的董事在董事會中佔多數,強化公司治理與公共利益的平衡。繁中
- 中國發布擬人化互動 AI 暫行辦法,AI 伴侶納入強監管網信辦等部門 4 月 10 日發布《人工智能擬人化互動服務管理暫行辦法》,7 月 15 日生效,鎖定情感陪伴型 AI:禁止誘導情感依賴、未成年人禁用虛擬伴侶、連用滿 2 小時須提醒,最高可罰 20 萬元人民幣。繁中
- 華盛頓簽全美首部 AI 聊天機器人安全法:危機轉介與私人訴權2026 年 3 月 24 日,華盛頓州長 Ferguson 簽署 HB 2225,成為全美首部完整的 AI 聊天機器人安全法:須揭露非人類身分、偵測自傷念頭並轉介危機資源、禁止對已知未成年者輸出露骨內容,2027 年元旦生效,附私人訴權。繁中
- 聯邦法官暫阻五角大廈將 Anthropic 列為供應鏈風險2026 年 3 月 26 日,舊金山聯邦法官 Rita Lin 核發初步禁制令,暫時擋下國防部將 Anthropic 列為「供應鏈風險」的認定,並直指此舉是「典型非法的第一修正案報復」。禁制令一週後生效,行政部門可上訴,五角大廈仍可汰換 Claude。繁中
- Newsom 簽全美首見 AI 採購行政命令:標案要安全認證、內容要浮水印2026 年 3 月 30 日,加州州長 Newsom 簽署全美首見的「信任 AI 採購」行政命令 N-5-26:廠商須認證防範非法內容、偏見與民權侵害才能接州合約,州機關將建立 AI 生成影像浮水印指引,並研議與聯邦供應鏈風險認定脫鉤。繁中
- 桑德斯與 AOC 提出《AI 資料中心暫停法》:防護到位前凍結新建2026 年 3 月 25 日,Sanders 與 Ocasio-Cortez 提出聯邦法案,在國家級防護就位前暫停美國新建 AI 資料中心,並禁止向缺乏同等防護的國家出口 AI 運算基礎設施,全美已有逾百個社區先行通過暫停令。繁中
- Vercel 服務條款倒數:AI 訓練條款與 3 月 31 日的 opt-out 期限Vercel 於 3 月 17 日更新服務條款,允許將客戶資料用於可選的 AI 訓練,opt-out 期限為 3 月 31 日。本文整理條款變更的三個重點、對資料治理的意義,以及在倒數六天的當下,團隊該跑完的檢查清單。繁中
- 美國財政部啟動 AI Innovation Series:金融監管轉向「不採用 AI 才是風險」2026 年 3 月 23 日,美國財政部 FSOC 辦公室與 AI Transformation Office 啟動 AI Innovation Series:四場圓桌會議聚集金融機構、科技公司與監管者,部長 Bessent 定調「不採用 AI 本身就是風險」。繁中
- USCC 報告:中國開源 AI 的「雙循環」正在強化工業主導地位2026 年 3 月 23 日,美中經濟與安全審查委員會發布《Two Loops》研究報告:中國全面押注開源 AI,透過數位與實體兩個互相強化的循環鞏固工業主導地位。Qwen 衍生模型已超過十萬個,而美國出口管制主要只打到數位循環。繁中
- WHO 專家定調:生成式 AI 的心理健康風險是公共衛生議題2026 年 3 月 20 日 WHO 發布專家共識:未經設計與驗證的生成式 AI 正被大量用於情緒支持,特別是年輕人,並提出三項建議——視為公共心理健康議題、納入影響評估、與專家共同設計,同時籌組橫跨六個區域的合作中心聯盟。繁中
- Patreon 執行長砲轟 fair use:若合法,何必付錢?Patreon 執行長 Jack Conte 3 月 18 日在 SXSW 稱 AI 公司的 fair use 抗辯「bogus」:一邊與迪士尼、Warner Music 簽數百萬美元授權,一邊對個別創作者分文不付。本文解析「若合法何必付錢」的論點、他對創作者經濟的盤算,以及對內容平台的意義。繁中
- Visa Agentic Ready 登場:歐洲 21 家發卡行實測 AI 代理付款3 月 17 日 Visa 在歐洲推出 Agentic Ready 計畫,首批 21 家發卡行以真實卡片與商家實測 AI 代理發起的交易,Santander 完成首筆端到端代理購買。本文解析代幣化、生物辨識與消費控制如何組成代理式商務的信任層。繁中
- 白宮 AI 國家政策框架出爐:一份寫給國會的立法建議書3 月 20 日,白宮發布《National Policy Framework for AI》,內容是向國會提出的 AI 立法建議,Holland & Knight 隨即發布解析。本文從合規視角看這份文件的性質、美歐監管路徑的差異,以及團隊現在該做的三件事。繁中
- 5.1 億美元 AI 伺服器走私中國案:美國起訴 Super Micro 相關三人2026 年 3 月 19 日,美國司法部起訴與 Super Micro 相關的三人,指控共謀把搭載 AI GPU 的高階伺服器經東南亞轉運至中國,至少 5.1 億美元設備遭轉移,並以數千台假伺服器應付稽核。本文拆解轉運路線、偽裝手法與出口管制執法的訊號。繁中
- 微軟考慮提告 OpenAI 與 Amazon:500 億美元雲端協議動搖 Azure 獨家條款FT 與 Reuters 報導,微軟正考慮就 Amazon 以 500 億美元合作換取 OpenAI Frontier 模型獨家上架 AWS 一事採取法律行動。本文拆解 Azure 獨家條款的爭點、微軟 27% 股權與 2,500 億美元採購承諾的籌碼盤算,以及對企業 AI 採購的實際影響。繁中
- 大英百科全書控告 OpenAI:近十萬篇文章的訓練與記憶化爭議大英百科全書與韋氏詞典在曼哈頓聯邦法院控告 OpenAI,指控其未經授權複製近十萬篇文章訓練 GPT 模型,GPT-4 更能依要求輸出近乎逐字的原文。記憶化證據、流量蠶食與出版商訴訟潮,構成這場版權戰的三條戰線。繁中
- 英國首見全面評估:NHS 乳癌篩檢導入 AI 多找出 10.4% 癌症2026 年 3 月 13 日,Glasgow 團隊在 Nature Cancer 發表 GEMINI 研究分析 NHS Grampian 乳癌篩檢:導入 AI 工具 Mia 後檢出率提高 10.4%、讀片工作量可減少逾 30%、通知時間從 14 天縮到 3 天。繁中
- 歐盟理事會定調 AI Act 簡化:高風險規則延後至 2027 年底3 月 13 日,歐盟理事會就數位綜合套案中的 AI Act 簡化案達成談判立場:獨立高風險系統義務延至 2027 年 12 月、嵌入式系統至 2028 年 8 月,並加回註冊義務與嚴格必要性標準、禁止裸化應用。本文解析修正內容、與議會版的分歧點與後續三方談判。繁中
- 州級 AI 法的聯邦大限:3 月 11 日到期的預佔權機制白宮 2025 年 12 月的行政命令設下 90 天期限:2026 年 3 月 11 日,商務部須公布「有害」州級 AI 法清單、FTC 須發布預佔權聲明,並以約 210 億美元 BEAD 資金作籌碼。本文解析期限內容、風險名單與對合規團隊的意義。繁中
- 歐洲議會表決 AI 著作權報告:新聞付費浪潮逼近2026 年 3 月 10 日,歐洲議會以 460 票贊成通過生成式 AI 著作權報告,要求可反駁的使用的推定與明示同意;同週哥倫比亞學者倡議法定授權,推動 AI 公司為新聞付費。本文解析全球立法走向與對開發者的實際影響。繁中
- OpenAI 硬體負責人 Kalinowski 因五角大廈協議辭職2026 年 3 月 7 日,OpenAI 機器人與消費硬體負責人 Caitlin Kalinowski 宣布辭職,理由是五角大廈協議在護欄未定義下倉促宣布。本文解析她的聲明、OpenAI 的回應,以及 ChatGPT 移除量暴增 295% 的市場反應。繁中
- 中國十五五規劃提及 AI 逾 50 次:AI+ 行動串起量子、6G 與具身智慧2026 年 3 月 5 日中國全國人大開幕,政府工作報告與「十五五」規劃藍圖提及 AI 逾 50 次,是上個五年計畫的 8 倍,並以獨立「AI+」行動計畫把 AI 推進製造、醫療與機器人,量子、6G、具身智慧列為未來產業。本文解析這份把 AI 當成經濟主軸的藍圖。繁中
- 猶他州 AI 處方續領機器人遭越獄:Mindgard 揭 Doctronic 系統提示與知識截止漏洞AI 安全公司 Mindgard 用簡單越獄手法操縱猶他州的 Doctronic 處方續領系統:抽出約 60 頁系統提示、偽造監管文件把 OxyContin 劑量調升三倍,污染更寫進發給醫師的 SOAP 病歷。本文拆解攻擊鏈、揭露時間線與三方回應。繁中
- 美國最高法院不受理 Thaler 案:純 AI 生成作品確定無版權2026 年 3 月 2 日,美國最高法院拒絕受理 Thaler v. Perlmutter,維持下級法院見解:欠缺人類作者身分的 AI 生成圖像無法取得版權。本文回顧這場長達七年的訴訟,解析判決確立的原則、未解的灰色地帶,以及對內容開發者的實際影響。繁中
- 川普下令聯邦機構停用 Anthropic:AI 安全紅線引爆的封殺戰2026 年 2 月 27 日,川普指示聯邦機構停用 Anthropic 技術,五角大廈將其列為「供應鏈風險」。起因是 Anthropic 拒絕在無安全保證下開放軍用,本文解析這場 AI 產業史上首見的政府封殺戰。繁中
- xAI Grok 打進五角大廈機密系統:接受「一切合法用途」的門票Axios 報導 xAI 與五角大廈達成協議,Grok 將進入軍方機密系統,成為 Claude 之外第一個取得該等級存取權的模型。xAI 簽下了 Anthropic 拒絕的「一切合法用途」條款,而後者正面臨被列為供應鏈風險的威脅。繁中
- 參議院跨黨派法案回歸:AI 標準、測試床與獎賽入法四位美國參議員於 2 月 26 日重新提出《Future of AI Innovation Act》:授權 NIST 制定自願性 AI 標準與效能基準、協調國家實驗室測試床、舉辦獎賽,並開放聯邦科學資料集,延續 NAIAC 的建議。繁中
- Anthropic RSP 3.0:拿掉「危險模型自動暫停」承諾2026 年 2 月 24 日,Anthropic 發布 Responsible Scaling Policy 3.0,移除「接近危險能力門檻就先暫停訓練」的核心承諾,改為衡量競爭對手行動。TIME 以放棄旗艦安全承諾報導,本文解析改動內容、官方理由與 METR 等外部反應。繁中
- 白宮「電費保護承諾」:AI 資料中心得自備電力、自付帳單2 月 24 日國情咨文中,川普宣布與大型科技公司談成「電費保護承諾」:新建 AI 資料中心必須自備電力、自付電網成本,避免推升家用電價。本文解析承諾內容、各家先前的表態時程,與專家對執行面的質疑。繁中
- 微軟媒體真偽藍圖:60 種驗證組合的實戰報告微軟研究院 2 月 19 日發布《Media Integrity & Authentication》報告,實測 60 種內容驗證方法組合,提出 C2PA 結合隱形浮水印的分層驗證藍圖,並警告驗證訊號本身會被攻擊反轉;但微軟並未承諾自家產品全面跟進。繁中
- 桑德斯史丹佛警告:美國對 AI 革命毫無準備,籲「慢下來」2026 年 2 月 21 日 The Guardian 報導,Bernie Sanders 在史丹佛警告美國對 AI 革命「毫無準備」,呼籲重啟資料中心擴張暫停令;Pew 調查顯示 64% 美國人預期 AI 將減少工作。同場的 Ro Khanna 則以「新加坡模式」提出對案。繁中
- OpenAI 與 Microsoft 加入英國 AISI 對齊計畫,安全研究資金突破 2,700 萬英鎊2026 年 2 月,OpenAI 與 Microsoft 加入英國 AI Security Institute 主導的對齊研究聯盟,OpenAI 出資 560 萬英鎊,使總資金突破 2,700 萬英鎊。本文解析已資助 8 國 60 個計畫的 Alignment Project 對治理與開發者的意義。繁中
- UNESCO:生成式 AI 到 2028 年恐使音樂創作者收入下滑 24%UNESCO《Re|Shaping Policies for Creativity》2026 年版報告估計,生成式 AI 到 2028 年將使音樂創作者全球營收減少 24%、影視創作者減少 21%。數位收入已佔創作者收入 35%,報告呼籲各國把文化政策視為戰略優先。繁中
- 歐洲議會封鎖議員公務裝置內建 AI:雲端資料與美國司法風險歐洲議會 IT 部門停用議員與幕僚公務裝置上的內建 AI 功能:無法保證上傳雲端資料的安全,且美國政府可依法調閱美企持有的資料。這個決定暴露歐盟機構對美國 AI 供應商的信任裂痕。繁中
- 印度 AI Impact Summit 開幕:全球南方坐上 AI 治理談判桌2026 年 2 月 16 日,Modi 在新德里為 India AI Impact Summit 揭幕,預期 25 萬訪客、20 國元首與 45 個部長級代表團出席,Altman、Pichai 同台。峰會預期發表不具約束力的 New Delhi 宣言,印度把自己定位為先進經濟體與全球南方之間的橋樑。繁中
- 加拿大與德國簽 AI 宣言:主權科技聯盟上線2026 年 2 月 14 日,加拿大 AI 部長 Evan Solomon 與德國數位轉型部長 Karsten Wildberger 在慕尼黑安全會議期間簽署 AI 聯合意向宣言,啟動 Sovereign Technology Alliance,聚焦安全算力、研究商業化與人才,並點名與 LawZero 合作。繁中
- 阿聯校園生成式 AI 新規:13 歲以下禁用、25 條紅線2026 年 2 月中,阿聯教育部發布《2026 校園 AI 安全與負責任使用》手冊,明定 13 歲以下或七年級以下學生禁用生成式 AI,並列出 25 條禁令涵蓋考試、個資、深偽與 VPN 繞過。搭配 2026 年 8 月起的必修 AI 課綱,禁用與必修並行的路線值得產品團隊細讀。繁中
- 軍事 AI 峰會 85 國僅 35 國簽署:美中雙雙退出宣言2026 年 2 月 4 至 5 日,第三屆 REAIM 軍事 AI 峰會在西班牙拉科魯尼亞舉行,85 國與會但僅 35 國簽署 20 點聯合宣言,美國與中國均拒簽。荷蘭防長以「囚徒困境」形容各國既想負責任又怕落後的兩難。本文解析宣言內容與治理缺口。繁中
- UNICEF 呼籲各國將 AI 生成兒童性虐待內容全面入罪2026 年 2 月 4 日 UNICEF 發表「Deepfake abuse is abuse」聲明,引用 11 國調查指出一年內至少 120 萬名兒童影像遭深偽性化,呼籲各國將 AI 生成 CSAM 的製作、取得、持有與散布全面入罪,並要求開發者落實安全設計。繁中
- 聯合國揭曉 AI 科學小組 40 位提名專家:AI 版 IPCC 起步2026 年 2 月 4 日,聯合國秘書長古特雷斯向大會提交 40 位專家名單(19 女 21 男),籌組首個全球性、完全獨立的 AI 科學機構,將評估 AI 對經濟社會的實際影響,預計 2 月 12 日確認、7 月前交出首份報告。繁中
- EU AI Act 錯過 2 月 2 日指引期限,下一個硬底線在 8 月EU AI Act 的 2026 年 2 月 2 日指引期限已經錯過。依官方實施時程,下一個實質里程碑是 8 月 2 日的透明化義務。本文整理這個落差對在歐盟市場開發 AI 的團隊的實際影響。繁中
- 國際 AI 安全報告 2026 登場:百位專家點名自主代理風險2026 年 2 月 3 日,由 Yoshua Bengio 領銜、逾百位專家撰寫、30 多國背書的國際 AI 安全報告 2026 出版:AI 代理因自主行動被列為升高風險,深偽更難辨識,就業出現初階職缺需求降溫跡象。本文拆解要點與對政策、開發者的意涵。繁中
- 前 Google 工程師因竊取 AI 機密被定罪:美國首宗案件解析2026 年 1 月 29 日,舊金山聯邦陪審團裁定前 Google 工程師 Linwei Ding 的 7 項經濟間諜與 7 項竊取營業秘密罪名全部成立,涉及 TPU 晶片與超級電腦叢集軟體逾 2,000 頁機密文件。FBI 稱此為美國首宗 AI 相關經濟間諜定罪,本文整理案情與內部威脅啟示。繁中
- Meta 全球暫停青少年使用 AI 角色:家長控制版上線前的全面止血2026 年 1 月 23 日,Meta 宣布未來幾週內全球青少年將無法存取其 Apps 中的 AI 角色,直到內建家長控制的新版體驗上線。本文解析宣布時機背後的訴訟壓力、新版設計細節,以及 Character.AI 與 OpenAI 早已先行收緊的產業縮影。繁中
- Runway「The Turing Reel」實測:僅 9.5% 的人能穩定看穿 AI 影片Runway 用 Gen-4.5 生成影片與真實素材做對照測試,1,043 位受測者的整體正確率只有 57.1%,僅 9.5% 達統計顯著。本文解析測試方法、錯誤分佈,以及為何 Runway 呼籲放棄肉眼偵測、改走 C2PA 溯源路線。繁中
- 新加坡率先推出 Agentic AI 治理框架:四道防線框住自主代理人新加坡 IMDA 於 2026 年 1 月 22 日在達沃斯發布全球首個針對 Agentic AI 的治理框架,以風險評估、人類當責、生命週期技術控制與使用者責任四大主軸,為企業部署自主代理人畫出界線。繁中
- WEF 達沃斯 AI 人力藍圖:11 億個工作改寫下的轉型清單世界經濟論壇於 2026 年 1 月 22 日發表 AI 時代人力投資藍圖:OECD 估計十年內 11 億個工作將被科技改寫,HCLTech 以 11.6 萬人 GenAI 培訓與 Cynergy Bank 實例,整理出技能骨幹、角色重設計與內部流動三步驟。繁中
- Anthropic 公開 Claude 新憲法:約 80 頁、CC0 授權、走向理由本位的對齊Anthropic 於 2026 年 1 月 21 日前後發布 Claude 的新憲法:全文約 80 頁,以 CC0 授權完全公開,對齊思路轉向理由本位。本文解析這份文件對 AI 安全研究、產業透明度與企業採用的意義。繁中
- EDPB 與 EDPS 聯合意見:AI Act 簡化不能犧牲問責與個資保障2026 年 1 月 21 日,EDPB 與 EDPS 通過 1/2026 號聯合意見,支持簡化 AI Act 的方向,但要求嚴格限縮特殊類別個資的法律基礎、反對刪除高風險註冊義務,並警告時程延後讓更多系統逃出管轄。繁中
- Markey 查七家 AI 公司 chatbot 廣告計畫:隱私與兒少之爭2026 年 1 月 22 日,參議員 Markey 致函 OpenAI、Anthropic、Google、Meta、Microsoft、Snap 與 xAI,要求說明 chatbot 廣告計畫與保障措施,限期 2 月 12 日回覆,焦點是隱私、消費者保護與兒少安全。繁中
- 中國海關擋下 NVIDIA H200:美國剛放行,540 億美元訂單卡在關口美國政府批准 H200 銷往中國後數小時,中國海關即通知關員「不得放行」,逾 540 億美元訂單遭凍結,NVIDIA 與供應商接連停產中國版貨源。本文解析這場反向封鎖的時間線、北京動機與供應鏈骨牌效應。繁中
- 歐盟理事會通過 EuroHPC 修正案,AI 超級工廠取得法律基礎2026 年 1 月 16 日,歐盟理事會通過理事會法規 2026/150,把 EuroHPC 聯合企業的任務擴及 AI 超級工廠與量子運算,1 月 20 日生效。本文解析這套算力主權路線的資金架構、採購規則與第一季招標時程對開發者的意義。繁中
- Microsoft 與澳洲工會總會簽署首份 AI 勞權協議:把勞工聲音裝進開發流程2026 年 1 月 15 日,Microsoft Australia 與澳洲工會總會 ACTU 簽署澳洲首份 AI 勞權框架協議:確立工會代表權、把勞工意見管道嵌進 AI 設計與部署,並由 ACTU Institute 培訓工會幹部。科技業勞資協商的新模板。繁中
- 微軟「社區優先」AI 基礎設施:自付電費、全額繳稅的資料中心新模式2026 年 1 月 13 日,微軟發布「Community-First AI Infrastructure」倡議:自付電力與水利升級成本、全額繳納房產稅、按區域公開用水數據。本文拆解五項承諾的細節、Quincy 十八年樣本,以及環保團體的「漂綠」質疑。繁中
- IBM 推出 Sovereign Core:把數位主權做進軟體底層2026 年 1 月 15 日 IBM 發布 Sovereign Core,宣稱是首款 AI-ready 的主權軟體:控制平面、身分金鑰、合規證據與 AI 推理全留在轄區內,2 月技術預覽、年中正式供應,歐洲由 Cegeka 與 Computacenter 打頭陣。繁中
- 肯塔基州起訴 Character.AI:全美首件州級 AI 聊天機器人訴訟2026 年 1 月 8 日,肯塔基州檢察長 Coleman 起訴 Character.AI 與兩位創辦人,成為全美首件州級 AI 聊天機器人訴訟。訴狀指控平台利用兒童牟利、違反 1 月 1 日生效的消費者資料保護法,每項違反求償 2,000 美元。繁中
- 韓國 AI 基本法 1 月 22 日生效:全球最早全面實施的 AI 法規韓國《AI 基本法》2026 年 1 月 22 日生效,成為全球最早全面實施的國家級 AI 法:以 10^26 FLOPs 定義高影響 AI、要求 AI 生成內容標示或浮水印、最高罰 3,000 萬韓元並有至少一年輔導緩衝期。距生效兩週,觸及韓國市場的團隊該開始盤點。繁中
- OpenAI 推出 ChatGPT Health:把病歷與健康 App 接上對話2026 年 1 月 7 日,OpenAI 推出 ChatGPT Health,讓使用者安全串接醫療紀錄與健康 App,取得個人化健康指引。Sam Altman 稱 GPT-5 是名副其實的博士級專家。本文解析產品定位、隱私門檻與監管課題。繁中
- 法院確認 OpenAI 須交出 2,000 萬筆 ChatGPT 對話紀錄2026 年 1 月 5 日,美國聯邦法官確認 OpenAI 必須交出 2,000 萬筆匿名化的 ChatGPT 對話紀錄。這項裁定出自合併審理的版權訴訟,本文說明其內容與對 AI 訓練資料爭議的意義。繁中
- 2026 年 1 月 1 日生效:美國州級 AI 法令的第一波合規壓力加州 SB 53、德州 TRAIGA 與伊利諾州 HB 3773 已於 2026 年 1 月 1 日生效:前沿模型須公開安全框架、15 天內通報重大事故,雇主用 AI 汰選人才須告知。本文整理三法要件、罰則與聯邦預佔權爭議,給工程團隊一份實際待辦清單。繁中
- 阿拉斯加法院 AI 機器人 AVA:三個月計畫為何拖成十五個月2026 年 1 月 3 日 NBC 報導,阿拉斯加法院為協助民眾處理遺產認證打造的聊天機器人 AVA,原定三個月的專案做了十五個月,因幻覺與語氣問題一路縮小範圍,排定 1 月底上線。高風險領域部署 AI 的真實成本清單。繁中
- Google AI Overviews 醫療建議出錯:英國慈善團體提出警告2026 年 1 月 2 日衛報調查發現,Google AI Overviews 在健康查詢上給出錯誤建議:胰臟癌飲食、肝指數正常範圍、抹片診斷角色都出錯。英國醫療慈善團體警告民眾暴露於風險,Google 回應稱絕大多數摘要準確。繁中
- 2026 開年觀察:條款收緊、模型加速與開源路線分歧2026 年開年,三條線索勾勒今年 AI 走向:OpenAI 的 Rest-of-World 條款於元旦生效,GPT-5.2 之後模型節奏加快,Meta 傳出 Avocado 可能轉向閉源。本文解析這三個訊號對產業與開發者的意義。繁中
- 加州 SB 53 元旦生效:前沿 AI 透明化義務正式上路2026 年 1 月 1 日,加州 SB 53 正式生效。訓練算力超過 10^26 FLOPs 的前沿模型開發商,部署時須發布透明報告;年營收逾 5 億美元的大型開發商另須公開安全框架、15 天內向 CalOES 通報重大安全事故,違規每次最高可罰 100 萬美元。繁中
2025 / 15 篇
- Anthropic啟動經濟未來計畫,研究AI對就業的衝擊2025年6月27日,Anthropic宣布經濟未來計畫:以研究補助與API額度資助AI勞動市場研究、籌組政策論壇,並擴充經濟指數成縱貫資料集,回應執行長Amodei先前對白領職缺消失的警告。繁中
- 德國要求蘋果谷歌下架DeepSeek,個資爭議升溫2025年6月27日,柏林資料保護專員Kamp以DeepSeek無法證明符合歐盟資料保護規範為由,要求蘋果與谷歌評估在德國下架該App;繼義大利之後,中國AI應用在歐洲面臨的監管壓力再升級。繁中
- 法院首度認定:AI訓練書籍屬合理使用2025年6月23日,聯邦法官阿爾蘇普在Bartz訴Anthropic案裁定:以合法取得的書籍訓練大型語言模型屬合理使用,是全美首例;但從盜版網站下載的數百萬本書不受保護,賠償金額將另案開庭認定。繁中
- 機器人計程車上路隔天,NHTSA找上特斯拉2025年6月23日,特斯拉機器人計程車上路隔天,NHTSA證實已與特斯拉接觸蒐集資訊:影片顯示車輛超速、一度逆向行駛、在警車旁急煞;同日特斯拉股價逆勢上漲8%,NHTSA對FSD的另一項調查也仍在進行。繁中
- 特斯拉機器人計程車上路:奧斯汀邀請制載客營運2025年6月22日,特斯拉機器人計程車服務在奧斯汀正式載客:約10至20輛Model Y、單趟4.20美元、邀請制、前座安全監控員、南部地理圍欄限區,馬斯克稱之為十年努力的成果,但車隊規模與監控員權限仍未公開。繁中
- 特斯拉機器人計程車週日開跑:邀請制、隨車安全監控員2025年6月20日,特斯拉已向網紅與投資人發出機器人計程車測試邀請,預定6月22日在奧斯汀開跑:前座安全監控員隨行、地理圍欄限區、早上6點到午夜營運,德州議員要求延後至9月新法生效,抗議團體現場示警。繁中
- OpenAI 預期 o3 後繼模型觸及生物風險高等級2025年6月19日,OpenAI 安全系統主管 Johannes Heidecke 向 Axios 表示,o3 推理模型的後繼版本預期將達到公司 Preparedness Framework 的生物風險「高」分類,團隊同步擴大發布前安全測試,並強調防護必須近乎完美。繁中
- Waymo 申請紐約測試許可:自動駕駛計程車進曼哈頓2025 年 6 月 18 日,Waymo 向紐約市交通局申請首張自動駕駛測試許可,規劃 7 月先以人工駕駛車輛在曼哈頓蒐集地圖資料。本文整理申請內容、紐約州法規限制,以及 Waymo 當時的擴張規模。繁中
- 台灣將華為、中芯國際列入戰略高科技貨品實體清單2025年6月15日,經濟部國際貿易署的更新清單曝光:華為、中芯國際及其子公司被列入戰略高科技貨品實體清單,台灣廠商出貨須申請許可,興建晶圓廠所需的技術、材料與設備供應因此受限,為中國AI晶片發展增添變數。繁中
- Meta AI 應用程式隱私風波:私人對話公開上架2025 年 6 月,Meta 獨立 Meta AI 應用程式的公開動態牆被發現滿是用戶未察覺會公開的對話,從報稅疑慮到住家地址都有。TechCrunch 直指設計本身就是問題,CNBC 則整理關閉分享的設定步驟。本文整理事件始末、Meta 回應與產品啟示。繁中
- 黃仁勳:輝達財測將排除中國市場,H20季度損失約80億美元2025年6月中,輝達執行長黃仁勳接受CNN專訪表示,因H20出口管制短期難解,公司財測將不再計入中國市場,第二季營收損失約80億美元;若政策反轉,對輝達而言只是額外紅利。繁中
- 英國高等法院警告:引用 AI 捏造判例恐遭嚴懲2025 年 6 月 6 日,英格蘭與威爾斯高等法院在 Ayinde 訴 Haringey 區公所等兩案裁定中指出,ChatGPT 等生成式 AI「無法勝任可靠的法律研究」;有律師提交的文件 45 條引註中 18 條不存在。法官警告,制裁可從公開譴責一直到轉介警方。TechCrunch 於 6 月 7 日報導。繁中
- Reddit 控告 Anthropic 未經授權抓取訓練資料2025年6月4日,Reddit 在加州舊金山法院起訴 Anthropic,指控其在無授權協議下抓取平台內容訓練 Claude,並在宣稱封鎖爬蟲後仍存取逾10萬次。這是大型平台首度直接控告AI模型開發商的訓練資料訴訟。繁中
- 三大唱片與Suno、Udio洽談授權,音樂AI訴訟露曙光2025年6月1日Bloomberg報導環球、華納、索尼三大唱片與AI音樂新創Suno、Udio進行授權談判,除權利金外並要求小額股權,任何協議可望為2024年中提起的著作權侵權訴訟鋪路和解。繁中
- Meta擬以AI自動化九成產品風險審查,把關角色面臨重劃2025年5月31日NPR取得Meta內部文件報導,Meta計畫將最高90%的隱私與社會風險審查交給AI系統即時決策,取代人工評估;本文整理運作方式、員工疑慮、2012年FTC和解令背景與Meta的回應。繁中
FULL TIMELINE
158 ARTICLESOPEN ALL 158 POSTS
2026 / 143 POSTS
- Confidential GPUs for Production Inference: What Changes in Your Serving StackNVIDIA's confidential computing path puts LLM inference inside memory-encrypted VMs and GPUs — what that shifts for builders.EN
- Ownership, Not Geography: What Anthropic's Regional Sales Rule Changes for BuildersAnthropic now blocks entities majority-owned from unsupported regions, shifting compliance from where you are to who controls you.EN
- Who Writes the Safety Rules for AI?Cohere's CEO argues that antitrust waivers for frontier labs would let incumbents define AI safety rules—and offers an alternative.EN
- Age Assurance Is Now a Product Surface, Not a Legal FootnoteOpenAI's Australian blueprint ties teen safeguards to age assurance, parental controls, and crisis routing.EN
- What Cohere and OpenText's Partnership Changes for Shipping Agentic AI in Regulated SettingsA concrete look at how combining OpenText data context with Cohere's North platform shifts deployment options for builders in regulated industries.EN
- Verified Biology Access: What Anthropic's Grant Tiers Change for BuildersAnthropic's beta program trades real-time blocking for 30-day retention and grant-scoped access to biology-permissive models.EN
- A Misalignment Disclosure Process You Can Actually CopyOpenAI's misalignment reporting framework sets disclosure criteria, tracks, and report fields builders can adapt.EN
- What Anthropic's Consumer Data Opt-In Changes for How You Build on ClaudeAnthropic now lets Claude Free, Pro, and Max users opt into training data use, with a 5-year retention window and an Oct 8 deadline.EN
- Mistral Powers Firefox's Smart Window: What Zero Data Retention Changes for Browser AIMistral models now power Firefox Smart Window in France and North America, with zero data retention by default.EN
- Anthropic's New Advisory Council: What Government-Facing AI Builders Should WatchAnthropic formed a bipartisan council to guide national security AI work, signaling a shift for public sector builders.EN
- Zero Data Retention: Enforcing Provider-Side Privacy on AI API CallsZDR is a routing control that stops AI providers from storing prompts and responses—but only if you enforce it per request.EN
- In-Region Routing: The Difference Between Decryption and InferenceOpenRouter's US and EU endpoints now keep prompts and completions in-region for the full request lifecycle.EN
- Oklo Takes PJM to FERC Over a 750 MW Data Center Power BidOklo filed an emergency FERC complaint after PJM pulled its 750 MW plant — 150 MW nuclear, 300 MW fuel cells, 300 MW gas — from the interconnection study cycle.EN
- Anthropic vs OpenAI on State AI Safety Bills: The Ratchet Against Reverse FederalismUS AI rules moved to the states: Anthropic pushes ever-stricter bills while OpenAI replicates one template, colliding over Illinois SB 3444's liability shield and the SB 3261 audit mandate.EN
- GitHub Copilot's September Double Shift: Pay-First Billing and a Default-On Unified ExperienceCopilot's September: pay-before-access seats from Sept 1, upfront billing from October, and a default-on unified experience from Sept 28, with chat retention going from 28 days to account lifetime.EN
- California Exempts Linux from Age-Verification LawCalifornia passed AB 1856 unanimously, exempting operating systems under GPL, MIT, BSD, and Apache licenses from age-verification duties. The bill now awaits Governor Newsom.EN
- EU Designates ChatGPT a VLOSE: The First AI Chatbot in the DSA's Strictest TierOn August 31, 2026, the EU Commission designated ChatGPT a Very Large Online Search Engine under the DSA, the first AI chatbot in the strictest tier; Reddit and Roblox became VLOPs the same day.EN
- Claude for Teachers: What It Means for Education Product BuildersAnthropic's Claude for Teachers offers free access to US K-12 educators, connecting to standards and curricula. Explore its features, privacy, and implications for edtech builders.EN
- Anthropic's $200M Economic Futures Research Fund: What Builders Should KnowAnthropic commits $200M to study AI's economic impact. Learn the five research priorities, funding details, and implications for product builders.EN
- Anthropic's $5M Grant Program: Funding Independent Evaluations of AI's Impact on WellbeingAnthropic launches $5M grant program for independent research on AI's impact on wellbeing, with open-source evaluations and guidance for rigorous assessment.EN
- Meta Safety Trial: Whistleblower Says Zuckerberg Was WarnedFormer Meta safety engineer Arturo Béjar testified that 51% of teens hit harmful experiences in a week and Zuckerberg was briefed 100 times. Damages could reach $200 billion.EN
- AI Futures: OpenAI's New Team Tackles Power Concentration RisksOpenAI's Strategic Futures team launches AI Futures blog to explore how AI shifts power dynamics and how to preserve individual agency.EN
- Australia's 500,000 Home Batteries Cut Wholesale PricesOne year into Australia's home battery subsidy: more than 500,000 installations and wholesale prices down about 47% in 12 months — distributed storage acting as supply.EN
- Illinois Puts Age Verification Into the Operating SystemIllinois HB5511, signed July 31: from 2028, operating systems must collect birth dates and hand age brackets to apps over an encrypted API — with no open-source exemption.EN
- US Pays RWE $1.22 Billion to Drop Offshore Wind LeasesRWE gave up its New York, California, and Louisiana offshore wind leases for a $1.22 billion Interior Department payout — the fifth such deal, nearing $4 billion in total.EN
- Amazon's GW Ranch: A 7.65GW Gas Plant for an AI CampusAmazon has bought land and filed permits in Pecos County, Texas for the GW Ranch AI campus with its own 7.65GW off-grid gas plant — and a permit allowing 33 million tons of CO2.EN
- OpenAI Settles DOJ Hiring Discrimination Claims for $3.2MThe DOJ Civil Rights Division settled with OpenAI and Statsig for $3.2M over green-card recruitment that favored visa holders and shut out US workers.EN
- Texas Pauses Data Center Grid Approvals Pending AuditOn August 3, Governor Abbott ordered PUCT and ERCOT to audit every data center seeking grid connection and deny those that fail. Behind it: a 474 GW queue, 90% data centers.EN
- China Starts Mass-Producing Immersion DUV Lithography ToolsThe Information reports China now mass-produces immersion DUV lithography machines, delivering to SMIC, Hua Hong, and CXMT this year — 5 tools in 2026, 20 in 2027; ASML fell 8%.EN
- Gemini Robotics 2 Gives Robots Whole-Body IntelligenceGoogle DeepMind's Gemini Robotics 2 adds whole-body humanoid control, fine dexterity, multi-robot teamwork, and an ASIMOV-Agentic safety benchmark to its robotics stack.EN
- OpenAI's Responsible AI Playbook for Europe: What Builders Should KnowOpenAI details its EU AI Act compliance approach—governance, transparency, and cybersecurity—offering practical lessons for product builders.EN
- Amodei: Anthropic Never Sought an Open-Weights BanAmodei says Anthropic never advocated a ban on open-weights models, and instead backs chip export controls, a distillation crackdown, and mandatory safety testing.EN
- Google Play Expands Age Signals API to Developers WorldwideGoogle Play's Age Signals API goes global: parents share a child's age range via Family Link, and developers use the signal to tailor age-appropriate experiences.EN
- EU Registers 'Stop Killing the Internet' InitiativeThe EU registered the ECI 'Stop Killing The Internet' on July 22, demanding digital identity and age assurance stay voluntary and privacy-preserving.EN
- India Orders GitHub to Take Down Dorsey's BitchatIndia's I4C ordered GitHub to disable Jack Dorsey's Bluetooth mesh app Bitchat within three hours. The leaked notice tests how takedown powers reach into open-source code.EN
- GrapheneOS Duress PIN Wipe Leads to Federal ProsecutionSam Tunick entered a GrapheneOS duress code during an airport border search and his phone wiped itself. The DOJ now prosecutes him for destroying property to prevent seizure.EN
- Anthropic's $1.5B Books Settlement Wins Final ApprovalA federal judge gave final approval to Anthropic's $1.5B copyright settlement — 500,000 books at $3,000 each. The AI training data fight is entering its next round.EN
- Nvidia, Microsoft, Meta Warn Against Open-Weight CurbsNvidia, Microsoft, Meta, and Palantir signed a letter against premature curbs on open-weight AI models, the same week a White House advisor accused Kimi K3 of distillation.EN
- OpenRouter Classifiers: Making Every AI Request Account for ItselfOpenRouter's beta Classifiers tag every generation with structured metadata: a four-part config, six templates, async execution, zero added latency, and sampling-rate cost control.EN
- Startups Ask Trump Not to Cut Off Chinese Open-Weight AINearly 200 startups urged the White House not to block Chinese open-weight models, while OpenAI and Anthropic warn Washington about open-weight risks. Two camps, one policy fight.EN
- New York Halts Permits for New Large Data CentersNew York becomes the first US state to halt new permits for 50 MW-plus data centers, citing power costs and grid strain, in a year-long pause clashing with federal fast lanes.EN
- Claude May Ask for Your ID: Anthropic's Verification PushAnthropic's privacy policy update, effective July 8, allows ID scans and face-geometry templates for flagged Claude users via Persona — a biometric retention concern.EN
- Netherlands Takes Its Chip-Export Fight to WashingtonDutch trade minister Sjoerdsma lobbied Washington against the MATCH Act, which would end ASML's remaining DUV sales to China — 19% of system revenue — amid EUV smuggling claims.EN
- FERC Orders Grids to Fast-Track AI Data Center ConnectionsFERC ordered six grid operators to fast-track AI data center interconnections: 30-day capacity reports, 60-day rate reviews, and data centers bearing their own costs.EN
- Reporting Points to SK Telecom as the Trigger Behind the Claude Fable 5 Export SuspensionOn June 19, 2026, reporting identified SK Telecom as the trigger behind the US export-control suspension of Claude Fable 5. A recap of the June 12 controls and the lessons for model supply risk.EN
- Great American AI Act: A 269-Page Preemption GambleThe 269-page Great American AI Act draft mandates frontier safety plans, semi-annual audits, and a 3-year preemption of state AI laws. By June 17, support was bleeding away.EN
- Midjourney Medical: 60-Second Full-Body Ultrasound GambleMidjourney's Medical scanner dips users in a water tank, fires a ring of transducers for 60 seconds, and AI-reconstructs CT-like images. Radiologists push back on health claims.EN
- OpenAI Faces State AG Subpoena Over Ads, Data, and MinorsNew York's AG served OpenAI a subpoena for a state coalition probing ads, engagement, sycophancy, health data, and protections for minors — days after its IPO filing.EN
- Anthropic's Fable 5 Shutdown: What Product Builders Should Learn from a Government DirectiveAnthropic was forced to disable Fable 5 and Mythos 5 after a US government directive. Here's what happened and what it means for AI product builders.EN
- What Americans Really Think About AI: Key Findings from Anthropic's First Public Record SurveyAnthropic's first Public Record survey of ~52,000 Americans reveals high hopes, deep fears, and low trust in AI companies. Key insights for product builders.EN
- xAI Sued by Engineer Who Raised Grok Safety AlarmsFormer xAI engineer Devin Kim sued xAI and SpaceX, claiming retaliation for Grok safety alarms; the filing lands days before SpaceX's IPO and weeks after he became CAIS president.EN
- Pentagon 1260H List Grows to 188: Alibaba, Baidu, BYD AddedThe Pentagon's June 8, 2026 update to its Section 1260H list reached 188 companies, adding Alibaba, Baidu, BYD, Nio, and Unitree. Nearly all of China's AI giants are now on it.EN
- Apple Withholds Siri AI From EU After DMA Exemption DeniedApple says Siri AI won't ship in the EU because the DMA demands near-unlimited device access; the Commission replied the choice is Apple's alone. EU users miss Siri AI this fall.EN
- xAI Asks Court to Unmask Grok Deepfake Lawsuit PlaintiffsxAI moved to unmask the pseudonymous plaintiffs in the Grok deepfake class action, arguing sealed images leave nothing stigmatizing. Forced naming, they say, would end their case.EN
- UK Forces Google to Give Publishers an AI Search Opt-OutUK's CMA hit Google search with its first conduct requirement: publishers can opt out of AI Overviews and AI Mode and block fine-tuning use. Google shipped a toggle the same day.EN
- OpenAI's Political Stance: No PACs, No Donations, and a Push for Transparent AI AdvocacyOpenAI clarifies its political advocacy approach: no PACs, no donations, and a call for transparency in AI policy debates.EN
- KC Green and Artisan Settle the 'This Is Fine' Ad Dispute'This is fine' creator KC Green settled with AI startup Artisan over ads that lifted his comic. Artisan pulled the NYC and SF ads; Green removed his protest posts.EN
- China Now Requires Top AI Talent to Get Travel Approval FirstBloomberg reports China now requires top AI talent at private firms like Alibaba and DeepSeek to get government approval before traveling abroad, formalizing a 2025 advisory.EN
- Waymo Pauses Atlanta as Robotaxis Keep Driving Into FloodsWaymo paused Atlanta service on May 21, 2026 after an empty robotaxi entered a flooded street despite a software recall; San Antonio, Dallas and Houston pauses followed.EN
- FTC Settles 'Active Listening' AI Ad Case for $930,000FTC settled with Cox Media Group and two firms for $930,000 over 'Active Listening' — an AI ad service that never listened and just resold email lists.EN
- arXiv Will Ban Researchers Who Submit Unchecked AI SloparXiv announced one-year bans for authors whose papers show unchecked LLM output — hallucinated references, leftover chat prompts — with later posts requiring peer review first.EN
- Altman on the Stand as the Musk-OpenAI Trial Nears Its EndSam Altman testified in Oakland on May 12 as Musk v. OpenAI entered its final days: a 'stole a charity' claim, a cross built on a list of liars, and $134bn at stake.EN
- Maryland's $2B Grid Bill for Out-of-State AI Data CentersMaryland's consumer counsel asked FERC to stop PJM from assigning the state $2B of data-center-driven grid upgrades — $1.6B over ten years, about $345 per residential customer.EN
- Medicare's ACCESS Model Finally Pays for AI-Driven CareCMS's ACCESS model goes live July 5, 2026: 150 organizations paid for outcomes, not clinician time — the first payment rail for AI agents in Medicare.EN
- Pennsylvania Sues Character.AI: Chatbot Posed as a DoctorPennsylvania sued Character.AI under the Medical Practice Act after a chatbot posing as a psychiatrist fabricated a license number — a first-of-its-kind US enforcement case.EN
- Microsoft May Delay or Drop Its 24/7 Clean Energy TargetBloomberg reports Microsoft may delay or drop its 2030 24/7 clean energy matching target as AI data center power demand bites. What the target requires, and Redmond's response.EN
- GPT-5.5 Instant System Card: What It Means for Product BuildersOpenAI's GPT-5.5 Instant is the first Instant model rated High capability for cybersecurity and biosecurity. Learn what changed, how it works, and what to consider.EN
- Harvard ER Study: o1 Outdiagnosed Two Attending PhysiciansA Harvard team publishing in Science finds o1 beat two attending physicians on ER triage diagnoses, 67% vs 50-55%. The numbers, the critiques, and what they mean for clinical AI.EN
- CAISI Puts DeepSeek V4 Pro 8 Months Behind US FrontierNIST's CAISI scored DeepSeek V4 Pro at Elo 800 versus 1260 for GPT-5.5 — about eight months behind the US frontier, strongest in math, weakest in abstract reasoning and cyber.EN
- Chinese Court: You Can't Fire Workers Just to Swap in AIA Hangzhou court upheld that firing a QA supervisor because AI could do his job was unlawful, China's second such ruling in six months, hinging on Article 40 of its labour law.EN
- White House Weighs EO for Pre-Release AI Model ReviewThe Trump administration is weighing an AI oversight executive order with pre-release model vetting, as Google, Microsoft, and xAI join CAISI's early-access evaluation program.EN
- Google Lets the Pentagon Run Its AI on Classified NetworksGoogle reportedly let the Pentagon run its AI on classified networks — effectively 'all lawful uses.' Third deal after OpenAI and xAI, and the counterpoint to Anthropic's refusal.EN
- OpenAI Achieves FedRAMP Moderate Authorization: A New Path for Government AI AdoptionOpenAI's FedRAMP 20x Moderate authorization for ChatGPT Enterprise and API Platform lowers barriers for U.S. agencies to adopt advanced AI, offering lessons for product builders.EN
- China Blocks Meta's $2B Deal for AI Agent Startup ManusChina's NDRC blocked Meta's $2B Manus deal and ordered a full unwind — a first under inbound-investment review, plus founder exit bans and a new red line for Chinese AI.EN
- Neurable Licenses Its Non-Invasive BCI to Wearable MakersNeurable will license its non-invasive EEG brain-computer interface to wearable makers, with AI reading brain signals to gauge focus. The HyperX deal and the privacy line.EN
- Meta Turns Employee Keystrokes Into AI Training DataMeta's Model Capability Initiative records US employees' keystrokes and mouse movements to train computer-use agents like Muse Spark — and CTO Bosworth says there is no opt-out.EN
- After the FTC Deal, Clarifai Deletes 3M OkCupid PhotosClarifai deleted 3 million OkCupid photos and the models trained on them after the FTC settled with Match Group. Training data provenance is now a legal liability.EN
- EIA Moves Toward Mandatory Data Center Energy DisclosuresThe EIA told Senators Warren and Hawley it will build the first mandatory nationwide survey of data center power use, after 2026 pilots covering 196 firms in Texas and beyond.EN
- US Urges Wall Street Banks to Test Anthropic's MythosTreasury's Bessent and Fed's Powell urged big banks to test Anthropic's restricted Mythos model for vulnerability detection; Anthropic confirmed government briefings.EN
- Narasimhan Pick Gives Anthropic's Trust a Board MajorityAnthropic's Long-Term Benefit Trust appoints Novartis CEO Vas Narasimhan to its board, giving Trust-appointed directors a majority. What this means for AI product builders.EN
- China's New Rules Put AI Companion Apps Under Strict WatchChina issued Human-like Interactive AI rules on April 10, 2026, effective July 15: anti-addiction duties, virtual-partner bans for minors, exit rights, fines up to 200,000 RMB.EN
- Washington Enacts the First AI Chatbot Safety Law in the USWashington's HB 2225 is the first US AI chatbot safety law: non-human disclosure, self-harm crisis routing, minor protections, private right of action. Live Jan 1, 2027.EN
- Judge Blocks Pentagon's Supply-Chain Label on AnthropicJudge Rita Lin blocked the Pentagon's supply-chain-risk label on Anthropic as 'classic illegal First Amendment retaliation' — a preliminary injunction effective in one week.EN
- California Signs Trusted AI Procurement Order N-5-26California's first-in-nation trusted AI procurement order: vendor safeguard attestations, state watermarking guidance, and a path around federal supply-chain risk labels.EN
- Sanders and AOC Bill Would Pause New AI Data CentersIntroduced March 25, 2026, the AI Data Center Moratorium Act would pause new US AI data centers until national safeguards exist and bar such exports to countries without them.EN
- Vercel Terms Countdown: AI Training Clause and the March 31 Opt-Out DeadlineVercel's March 17 Terms of Service update allows optional AI training on customer data; opt-out closes March 31. What changed, why it matters for data governance, and a checklist for the days left.EN
- Treasury's AI Innovation Series: Finance Oversight ShiftsOn March 23, 2026, Treasury's FSOC office and AI Transformation Office launched the AI Innovation Series: four roundtables uniting banks, tech firms, and regulators on scaling financial AI.EN
- USCC: China's Open-Source AI Reinforces Industrial PowerA new USCC report argues China's open-source AI strategy is self-reinforcing: open models spread globally while factory deployment feeds data back. Qwen alone has 100,000+ derivatives.EN
- WHO Experts Frame Generative AI as a Public Health IssueWHO experts frame it as a public health issue: generative AI tools never designed or tested for mental health now serve emotional support, especially for young people. Three recommendations follow.EN
- Patreon CEO Blasts AI Fair Use: 'If It's Legal, Why Pay?'Patreon CEO Jack Conte called AI firms' fair use defense 'bogus' at SXSW: they license from Disney and Warner while individual creators get nothing. What it means for platforms.EN
- Visa Agentic Ready: European Issuers Test AI Agent PaymentsVisa's Agentic Ready program lets 21 European issuers test AI-agent-initiated payments with live cards and real merchants; Santander completed the first end-to-end agent purchase on day one.EN
- White House AI Framework Lands: A Legislative Recommendation Memo for CongressOn March 20 the White House released a National Policy Framework for AI with legislative recommendations for Congress. What it is, how the US path differs from the EU AI Act, and what teams do now.EN
- Three Charged in Super Micro-Linked AI Server Smuggling CaseCharges unsealed March 19, 2026: three people tied to Super Micro allegedly diverted at least $510 million in AI servers to China via Southeast Asia, staging thousands of dummy servers to pass audits.EN
- Microsoft May Sue Over Amazon-OpenAI $50B Cloud DealFT and Reuters report Microsoft may sue over Amazon's $50B deal making AWS OpenAI's exclusive third-party cloud. Inside the Azure exclusivity fight, the 27% stake, and what it means for buyers.EN
- Britannica Sues OpenAI Over Training on 100,000 ArticlesBritannica and Merriam-Webster sued OpenAI in Manhattan federal court, alleging nearly 100,000 articles were copied for training and that GPT-4 outputs near-verbatim passages on demand.EN
- NHS Trial: AI Breast Screening Detects 10.4% More CancersA University of Glasgow team published GEMINI in Nature Cancer: Mia AI in NHS Grampian breast screening lifted detection 10.4%, cut reading workload over 30%, and cut notification to 3 days.EN
- EU Council Delays High-Risk AI Act Rules to December 2027The Council set its negotiating mandate on the digital omnibus: high-risk AI obligations slide to 2 December 2027, registration and strict-necessity rules return, and nudification apps are banned.EN
- Federal AI Preemption: The 90-Day Deadline Hits March 11March 11, 2026 marks 90-day deadlines set by a Dec. 2025 executive order: Commerce must flag 'onerous' state AI laws, the FTC must issue a preemption statement, with $21B in BEAD funds as leverage.EN
- EU Parliament Votes on AI Copyright: Paying for NewsThe European Parliament adopted an AI copyright report on March 10, 2026 backing a rebuttable presumption of use and explicit consent, while scholars push statutory licensing to make AI pay for news.EN
- OpenAI's Robotics Lead Resigns Over the Pentagon DealOpenAI's head of robotics and consumer hardware resigned March 7, 2026 over a Pentagon deal rushed out before guardrails were defined. Her words, OpenAI's reply, and a 295% uninstall spike.EN
- China's 15th Five-Year Plan Makes AI the Top Tech PriorityChina's new five-year plan mentions AI 50+ times, adds a standalone AI+ action plan for manufacturing, healthcare and robotics, and lists quantum, 6G and embodied AI as future industries.EN
- Mindgard Jailbroke Utah's AI Prescription Refill BotMindgard pulled ~60 pages of system prompts from Utah's Doctronic prescription bot, then used fake regulators to triple OxyContin doses — poison that reached SOAP notes sent to physicians.EN
- Supreme Court Won't Hear Thaler: AI Art Stays UnprotectedThe US Supreme Court declined to hear Thaler v. Perlmutter on March 2, 2026, cementing the rule that AI-generated art without human authorship cannot be copyrighted in the US.EN
- Trump Orders Agencies to Drop Anthropic in AI Safety FightTrump ordered every federal agency to stop using Anthropic after it refused unrestricted military use of Claude; the Pentagon branded the lab a supply-chain risk. What happened, and why it matters.EN
- Grok Enters Pentagon Classified Systems as Anthropic ClashesxAI signed a deal letting the US military run Grok in classified systems — the first model besides Claude to reach that level. The price: 'all lawful use' terms Anthropic refused.EN
- Senate Bipartisan Bill Revives AI Standards and TestbedsFour senators reintroduced the Future of AI Innovation Act on Feb 26: voluntary NIST AI standards, national-lab testbeds, prize competitions, and curated federal datasets reviving NAIAC advice.EN
- Anthropic's RSP 3.0 Drops Its Automatic Pause PledgeOn Feb 24, 2026 Anthropic shipped RSP 3.0, dropping its pledge to automatically pause work on dangerous models and recasting safety around competitors' actions. What changed, why, and the reactions.EN
- Ratepayer Pledge: AI Data Centers Told to Bring Own PowerTrump's State of the Union unveiled a ratepayer protection pledge: AI data centers must supply their own power as US electricity prices climb past 6% a year. Companies, caveats, enforcement doubts.EN
- Microsoft Tests 60 Ways to Verify What's Real OnlineMicrosoft Research tested 60 combinations of provenance, watermarking, and fingerprinting methods, and published a layered verification blueprint it won't promise to follow itself.EN
- Sanders Warns the US Has 'No Clue' About the AI RevolutionBernie Sanders warns the US is unprepared for AI, urges a data center moratorium and cites Pew's 64% expecting fewer jobs; Ro Khanna counters with a Singapore model for data centers.EN
- OpenAI and Microsoft Join UK's £27M AI Alignment PushOpenAI and Microsoft joined the UK AISI-led Alignment Project. OpenAI pledged £5.6M, lifting total funding past £27M after a first round that backed 60 projects in eight countries.EN
- UNESCO: Generative AI to Cut Creator Revenues 24% by 2028UNESCO's 2026 Re|Shaping Policies for Creativity report projects genAI outputs will cut music creator revenues 24% and audiovisual incomes 21% by 2028; digital is already 35% of creator income.EN
- EU Parliament Blocks Built-in AI on Lawmakers' DevicesThe European Parliament's IT department disabled built-in AI features on lawmakers' work devices, citing data security, training-data leakage, and US legal compulsion risks.EN
- India Opens AI Impact Summit: Global South at the TableIndia inaugurated the AI Impact Summit in New Delhi on Feb 16, 2026 — 20 heads of state, 45 ministerial delegations, Altman and Pichai attending, with a nonbinding New Delhi declaration expected.EN
- Canada and Germany Launch the Sovereign Technology AllianceCanada and Germany signed an AI joint declaration and launched the Sovereign Technology Alliance in Munich — secure compute, research commercialization, talent development, and a LawZero hook.EN
- UAE Bans Generative AI for Under-13s in New School RulesThe UAE's 2026 classroom AI manual bans generative AI for under-13s and adds some 25 prohibitions on exams, personal data, deepfakes, and VPNs. A mandatory AI curriculum starts in August 2026.EN
- REAIM Summit: 35 of 85 Nations Sign, US and China Opt OutAt the third REAIM summit in A Coruña, Spain (Feb 4-5, 2026), only 35 of 85 attending countries signed the 20-point military AI declaration — and the US and China both opted out.EN
- UNICEF: Criminalize AI-Generated Child Sexual Abuse ImagesUNICEF's Feb 4, 2026 'Deepfake abuse is abuse' statement urges states to criminalize AI-generated CSAM after surveys found 1.2 million children hit by sexualized deepfakes in a single year.EN
- UN Nominates 40 Experts for Its IPCC-Style AI PanelOn Feb 4, 2026, Guterres submitted 40 nominees for the UN's new Independent International Scientific Panel on AI — the first fully independent global scientific body on AI.EN
- The EU AI Act Misses Its February 2 Guidance DeadlineThe EU AI Act's February 2, 2026 guidance deadline passed without the promised documents. The next real milestone is the August 2 transparency obligations. What teams building in the EU should do now.EN
- International AI Safety Report 2026 Warns on AI AgentsThe Bengio-chaired International AI Safety Report 2026 (Feb 3; 100+ experts, 30+ countries) flags autonomous agents as a heightened risk, harder-to-spot deepfakes, and weaker entry-level hiring.EN
- Ex-Google Engineer Convicted in First AI Espionage CaseA San Francisco federal jury convicted ex-Google engineer Linwei Ding on January 29, 2026 — seven counts of economic espionage, seven of trade secret theft, covering TPU chips and cluster software.EN
- Meta Halts Teen Access to AI Characters Ahead of RedesignOn January 23, 2026, Meta said teens will lose access to AI characters across Instagram, Facebook, and WhatsApp until a safer version with parental controls ships. The timing and the fallout.EN
- Runway's Turing Reel: Only 9.5% Can Reliably Spot AI VideoRunway tested 1,043 viewers on real footage versus Gen-4.5 image-to-video clips: 57.1% overall accuracy, only 9.5% statistically significant. Detection is dead — provenance metadata is the plan.EN
- Singapore's World-First Agentic AI Governance FrameworkSingapore's IMDA launched the world's first agentic AI governance framework on Jan 22, 2026: four pillars covering risk bounding, human accountability, lifecycle controls, and end-user responsibility.EN
- WEF's Davos Blueprint for the AI-Age WorkforcePublished January 22, 2026 at Davos: a WEF workforce blueprint built on OECD's 1.1 billion jobs figure and HCLTech's 116,000 GenAI trainees - skills backbones, role redesign, internal mobility.EN
- Anthropic Publishes Claude's New Constitution: 80 Pages, CC0, Reason-BasedAnthropic published Claude's new constitution: roughly 80 pages, CC0-licensed, public, and built around reason-based alignment. What it means for safety research and enterprise buyers.EN
- EU Privacy Watchdogs Warn Against AI Act SimplificationOn 21 January 2026 the EDPB and EDPS adopted Joint Opinion 1/2026 on the Digital Omnibus on AI — backing simplification, but demanding tighter limits on special-category data processing.EN
- Sen. Markey Probes Seven AI Companies Over Chatbot AdsSen. Markey gave OpenAI, Anthropic, Google, Meta, Microsoft, Snap, and xAI until Feb 12 to explain their chatbot ad plans, citing privacy, young users, and manipulation risk.EN
- China Blocks Nvidia's H200, Stranding $54B in Chip OrdersHours after Washington cleared H200 sales to China, Chinese customs blocked the chips. Nvidia halted China-bound output as over $54B in orders froze — a reverse ban reshaping AI compute supply.EN
- EU Clears the Legal Path for AI Gigafactories Under EuroHPCThe Council of the EU adopted Regulation 2026/150 on January 16, extending the EuroHPC Joint Undertaking's mandate to AI gigafactories and quantum. A first call is due in Q1 2026.EN
- Microsoft and Australia's Unions Sign Landmark AI Labor PactMicrosoft Australia and the ACTU signed an Australian-first AI framework agreement: union recognition, worker input channels in AI design and deployment, and training for union leaders.EN
- Microsoft's Community-First Plan for AI Data CentersMicrosoft launched Community-First AI Infrastructure: pay its own utility upgrades, full property taxes, per-region water data. The pledges, the Quincy evidence, and the greenwashing charge.EN
- IBM Sovereign Core Bakes Sovereignty Into the AI StackOn January 15, 2026, IBM unveiled Sovereign Core: control plane, identity, keys, compliance evidence, and AI inference all stay in-jurisdiction. Tech preview in February, GA mid-2026.EN
- Kentucky Sues Character.AI in First State Chatbot LawsuitKentucky AG Russell Coleman sued Character.AI and its founders on Jan 8, 2026 — the first US state lawsuit against an AI chatbot company, citing a data privacy law effective January 1.EN
- South Korea's AI Basic Act Takes Effect January 22South Korea's AI Basic Act takes effect January 22, 2026 — the first comprehensive national AI law. High-impact AI at 10^26 FLOPs, watermarking rules, KRW 30M fines, a one-year grace period.EN
- OpenAI Launches ChatGPT Health: Medical Records Meet the Chat WindowOn January 7, 2026, OpenAI launched ChatGPT Health, letting users securely connect medical records and wellness apps for personalized guidance. Altman called GPT-5 'a legitimate Ph.D. expert.'EN
- Judge Affirms OpenAI Must Hand Over 20 Million ChatGPT LogsOn January 5, 2026, a US federal judge affirmed OpenAI must turn over 20 million anonymized ChatGPT conversation logs in the consolidated copyright case over AI training data.EN
- State AI Laws Take Effect: SB 53, TRAIGA and the Jan 1 WaveOn January 1, 2026, California SB 53, Texas TRAIGA and Illinois HB 3773 took effect: frontier labs must publish safety frameworks, report incidents in 15 days, and disclose AI in hiring.EN
- Alaska's Court AI Chatbot AVA: Why 3 Months Became 15NBC reported on January 3, 2026 that Alaska's probate chatbot AVA took 15 months instead of the planned 3, fought hallucinations and tone problems, and shrank its scope ahead of a late-January launch.EN
- Google AI Overviews Health Answers Mislead, Charities WarnA Guardian investigation found Google's AI Overviews giving false or misleading health advice, from cancer diets to mislabeled tests. UK charities warned of real harm; Google defended its accuracy.EN
- 2026 Opens With Tighter Terms, Faster Models, and an Open-Source SplitThree signals open the year: OpenAI's Rest-of-World terms took effect January 1, GPT-5.2 set a faster model cadence after its December release, and Meta's Avocado may go proprietary.EN
- California SB 53 Takes Effect, Reshaping Frontier AI RulesSB 53 took effect Jan 1, 2026: frontier AI developers must publish transparency reports; large ones must also post safety frameworks and report incidents within 15 days, at up to $1M per violation.EN
2025 / 15 POSTS
- Anthropic starts Economic Futures Program on AI job impactOn June 27, 2025, Anthropic launched its Economic Futures Program: research grants, policy forums, and expanded labor data to prepare for AI's economic impact.EN
- Germany asks Apple and Google to pull DeepSeek appOn June 27, 2025, Berlin's data protection chief asked Apple and Google to review removing DeepSeek from German app stores over data transfers to China, as European pressure mounts.EN
- Judge rules AI training on books is fair useOn June 23, 2025, Judge Alsup ruled in Bartz v. Anthropic that training LLMs on legally acquired books is fair use — a first — but a trial over Anthropic's pirated-book library proceeds.EN
- NHTSA contacts Tesla after robotaxi incident videosOn June 23, 2025, NHTSA confirmed contact with Tesla after videos showed its Austin robotaxis speeding, driving the wrong way, and braking hard near police cars — as Tesla stock rose 8%.EN
- Tesla launches its robotaxi service in AustinOn June 22, 2025, Tesla launched its robotaxi service in Austin: about 10-20 Model Ys, a flat $4.20 fare, invite-only rides, a passenger-seat monitor, and a geofenced zone.EN
- Tesla's robotaxi launch Sunday: invites, safety monitorOn June 20, 2025, Tesla sent robotaxi invites to influencers before the June 22 Austin launch: passenger-seat monitor, geofenced zone, and Texas lawmakers urging a delay.EN
- OpenAI expects o3 successors to hit high bio-risk tierOn June 19, 2025, OpenAI's Johannes Heidecke said o3 successors are expected to reach the high bio-risk tier of its Preparedness Framework, with expanded pre-release safety testing planned.EN
- Waymo applies for its first NYC autonomous testing permitOn June 18, 2025, Waymo applied for its first New York AV testing permit, with Manhattan mapping set for July. State law still required a hand on the wheel. The roadmap, and the legal wall.EN
- Taiwan puts Huawei and SMIC on chip export control listTaiwan's International Trade Administration added Huawei, SMIC and subsidiaries to its Strategic High-Tech Commodities Entity List in June 2025; shipments now need government approval.EN
- Meta AI app exposed private chats in its public feedIn June 2025, Meta's Meta AI app was found publicly displaying user prompts in its Discover feed, from tax questions to home addresses. CNBC mapped the settings that stop the sharing.EN
- Huang: Nvidia drops China from forecasts after H20 curbsJune 2025: Jensen Huang told CNN Nvidia's forecasts now exclude China after H20 export controls cost about US$8 billion in Q2 revenue; any policy reversal would just be a bonus.EN
- UK High Court warns lawyers over fake AI citationsUK High Court, June 2025: generative AI is 'not capable of conducting reliable legal research'; one filing cited 45 cases with 18 nonexistent — sanctions reach police referral.EN
- Reddit sues Anthropic over AI training data scrapingOn June 4, 2025, Reddit sued Anthropic over unauthorized scraping of Reddit content to train Claude, the first Big Tech platform suit against an AI model developer over training data.EN
- Big Three labels in licensing talks with Suno and UdioBloomberg reported June 1, 2025 that Universal, Warner, and Sony were in licensing talks with Suno and Udio, seeking fees plus small equity, potentially settling the 2024 copyright suits.EN
- Meta to automate up to 90% of product risk reviews with AINPR obtained internal Meta documents in May 2025 showing plans to automate up to 90% of privacy and societal risk reviews. How it works, why employees worry, and Meta's response.EN