[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-ai-regulation-maps-the-rules-you-need-zh":3,"article-related-ai-regulation-maps-the-rules-you-need-zh":30,"series-industry-27f4c1c1-c8a3-4f2f-8d0c-aa9199c40ab3":75},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":22,"views":26,"created_at":27,"published_at":28,"topic_cluster_id":29},"27f4c1c1-c8a3-4f2f-8d0c-aa9199c40ab3","ai-regulation-maps-the-rules-you-need-zh","AI 規範圖出你該守的規則","\u003Cp data-speakable=\"summary\">以前 AI 政策只剩空話，現在我把規範拆成你能直接照做的規則。\u003C\u002Fp>\u003Cp>我盯 AI 政策一陣子了。老實說，最煩的不是它變多，而是很多團隊還在用一張紙想解決所有事：寫個 policy、貼個 trust statement、交差。這招在 demo 時代還能混，產品一旦碰到招募、授信、醫療、內容審核，整個就會露餡。你以為自己在做治理，其實只是把風險包裝得比較像樣。\u003C\u002Fp>\u003Cp>真正讓我想通的，是我去看了 \u003Ca href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FRegulation_of_artificial_intelligence\">Wikipedia 的人工智慧規範頁\u003C\u002Fa>。它不是某家廠商的漂亮框架，反而因為夠雜，才把現場長什麼樣講清楚：硬法、軟法、各國規則、國際倡議，還有 trustworthy AI、responsible AI、ethical AI 這些常被混著用的字。很多時候，它們根本不是同一回事。\u003C\u002Fp>\u003Cp>我把這頁拆一拆，抓出對開發者真的有用的部分。不是空泛辯論，是你在寫產品、過法務、做 launch gate 時，會真的撞到的結構。你如果在台灣做 AI 產品，這篇就是給你一份可以直接抄的治理骨架。\u003C\u002Fp>\u003Ch2>AI 規範不是一本法典，是一堆壓力疊在一起\u003C\u002Fh2>\u003Cblockquote>“The regulatory and policy landscape for AI is an emerging issue in jurisdictions worldwide.”\u003C\u002Fblockquote>\u003Cp>白話翻譯就是：你找不到一本讀完就結案的 AI 法規大全。你面對的是地方性法律、產業法規、國際指引、標準、公司內規一起壓下來。只看單一 checklist，通常就是自己騙自己。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784882006925-6jc3.png\" alt=\"AI 規範圖出你該守的規則\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>我以前跟工程團隊講 AI 治理，他們最常問我一句：到底合不合法？我懂，這\u003Ca href=\"\u002Fnews\u002Fgemini-live-camera-turns-seeing-into-help-zh\">問題\u003C\u002Fa>很自然，但太乾淨了。現實比較像：哪個國家、哪個場景、哪個風險等級、哪個部門先看。這些條件一變，答案就變。\u003C\u002Fp>\u003Cp>Wikipedia 也提到像 \u003Ca href=\"https:\u002F\u002Fwww.oecd.org\u002F\">OECD\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fstandards.ieee.org\u002Findustry-connections\u002Fec\u002Fautonomous-systems.html\">IEEE\u003C\u002Fa> 這種不一定直接執法的機構，也在塑造 AI 該怎麼被看待。這代表規範不只有法律，還包括標準、原則、框架。你如果只看法條，會錯過很多在法條落地前就先定調的東西。\u003C\u002Fp>\u003Cp>實操寫法很簡單：不要找「一份 AI 政策」，改成做「政策堆疊」。\u003C\u002Fp>\u003Cul>\u003Cli>一層管產品風險分級。\u003C\u002Fli>\u003Cli>一層管各地法規差異。\u003C\u002Fli>\u003Cli>一層管內部審核與核准。\u003C\u002Fli>\u003Cli>一層管紀錄、稽核、事件處理。\u003C\u002Fli>\u003C\u002Ful>\u003Cp>這聽起來很官僚，我知道。但被法務追著問、被客戶稽核、被主管機關點名的時候，官僚至少還能救命。\u003C\u002Fp>\u003Ch2>軟法最容易被嫌煩，卻最早決定你會不會踩雷\u003C\u002Fh2>\u003Cblockquote>“Since 2016, numerous AI ethics guidelines have been published in order to maintain social control over the technology.”\u003C\u002Fblockquote>\u003Cp>翻譯一下就是：很多政府和機構會先丟原則、指南、守則，先把大家的期待框起來，再慢慢往硬法走。這些文件單獨看不一定有強制力，但它們會先變成大家講話的共同語言。\u003C\u002Fp>\u003Cp>頁面裡提到 \u003Ca href=\"https:\u002F\u002Fwww.partnershiponai.org\u002F\">Partnership on AI\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fstandards.ieee.org\u002Findustry-connections\u002Fec\u002Fautonomous-systems.html\">IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Foecd.ai\u002Fen\u002Fai-principles\">OECD AI Principles\u003C\u002Fa>。它們不是法條，但它們會影響採購問卷、企業客戶審查、風險部門怎麼問你問題。你如果看不懂這套語言，policy 寫得再漂亮也像沒上過場。\u003C\u002Fp>\u003Cp>我看過不少團隊把 soft law 當公關文案。結果一到 enterprise sales，對方問的每一題都剛好踩在那些原則上。更麻煩的是，出事後主管機關也常拿這些原則回頭問：你當時有沒有合理注意？\u003C\u002Fp>\u003Cp>實操寫法：把軟法當成早期預警系統。\u003C\u002Fp>\u003Cul>\u003Cli>先列出你對外宣稱的原則。\u003C\u002Fli>\u003Cli>對照客戶常要求的框架。\u003C\u002Fli>\u003Cli>把每個原則翻成一個控制項。\u003C\u002Fli>\u003Cli>寫清楚誰負責、怎麼測、多久重看一次。\u003C\u002Fli>\u003C\u002Ful>\u003Cp>像「透明」這種字不要只留在簡報裡。它至少要對應到 model card、使用者揭露、紀錄保存、以及可解釋流程。\u003C\u002Fp>\u003Ch2>硬法慢，但它真的決定你能不能上線\u003C\u002Fh2>\u003Cblockquote>“The European Union adopted in 2024 a common legal framework for AI with the AI Act.”\u003C\u002Fblockquote>\u003Cp>這句話的意思很直接：AI 規範已經從政策討論，走到可執行義務。EU AI Act 是最清楚的訊號之一。它不是唯一要看的一條，但很多團隊會拿它當參考座標，因為它把風險分級治理講得很具體。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784882002184-e2fw.png\" alt=\"AI 規範圖出你該守的規則\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>Wikipedia 也寫到，美國聯邦機關在 2024 年推出了 59 項 AI 相關規範，45 個州合計提出將近 700 項 AI 相關法案。這代表美國不是在等一部超大聯邦法一次收工，而是機關命令、州法壓力一起來。你的產品只要碰美國市場，就不能假裝那邊還沒開始動。\u003C\u002Fp>\u003Cp>這裡最常出事的是產品團隊。大家很會想 launch，不太會先想 classification。但規範看的是 use case，不是你模型名字叫什麼。拿來聊天的 bot 跟拿來做招募篩選、醫療分流，待遇本來就不一樣。\u003C\u002Fp>\u003Cp>實操寫法：建立 use-case inventory，不要只管 model inventory。\u003C\u002Fp>\u003Cul>\u003Cli>列出每個對外功能。\u003C\u002Fli>\u003Cli>寫清楚它影響什麼決策。\u003C\u002Fli>\u003Cli>標出上線地區。\u003C\u002Fli>\u003Cli>標記是否碰到權益、存取、或安全。\u003C\u002Fli>\u003C\u002Ful>\u003Cp>如果你只追模型版本號，很多真正的風險你根本沒記到。\u003C\u002Fp>\u003Ch2>真正該管的是責任歸屬，不只是準確率\u003C\u002Fh2>\u003Cblockquote>“AI governance... encompass[es] the questions of who is accountable for AI systems, what elements are governed, when governance occurs within the development lifecycle, and how it is implemented through frameworks, tools, or models.”\u003C\u002Fblockquote>\u003Cp>白話翻譯就是：治理不是哲學課，是責任地圖。系統出事時，不能只說「是供應商的模型」或「是 AI 團隊的事」。組織內部一定要有人對這個風險負責。\u003C\u002Fp>\u003Cp>Wikipedia 也提到，部署 AI 的組織在建立 trustworthy AI、遵守原則、承擔風險緩解責任上，扮演核心角色。這句很重要，因為它直接打掉一種常見偷懶：上游模型商不會替你扛下游後果。你只要把它放進產品，你就在鏈上。\u003C\u002Fp>\u003Cp>我自己最有感的是這件事。模型可能來自 \u003Ca href=\"\u002Ftag\u002Fapi\">API\u003C\u002Fa>，但 prompt、閾值、人工覆核、使用者看到的說法，全都是你自己的。只要使用者依賴這個結果，你就不可能假裝自己只是轉發別人的輸出。\u003C\u002Fp>\u003Cp>實操寫法：責任一定要寫進文件，不要只停在 Slack。\u003C\u002Fp>\u003Cul>\u003Cli>每個 AI 功能都指定 business owner。\u003C\u002Fli>\u003Cli>每個模型或工作流都指定 technical owner。\u003C\u002Fli>\u003Cli>高風險場景要有 legal 或 risk reviewer。\u003C\u002Fli>\u003Cli>補上事件、申訴、覆寫的升級路徑。\u003C\u002Fli>\u003C\u002Ful>\u003Cp>最好把這些責任放進 release 流程。能不能上線，不該等到出事後才問。\u003C\u002Fp>\u003Ch2>政策字眼會變，別拿漂亮詞當治理\u003C\u002Fh2>\u003Cblockquote>“Trustworthy AI”, “responsible AI”, and “ethical AI” have shifted in meaning over time and are often used interchangeably.\u003C\u002Fblockquote>\u003Cp>翻譯一下就是：政策文件裡的詞會老化，而且老得比產品快。兩年前看起來很精準的字，今天可能已經很空。問題是很多團隊超愛把這些詞直接貼進文件，卻從來沒定義。\u003C\u002Fp>\u003Cp>頁面引用 Charlotte Stix 的觀察，我覺得很準。你如果寫了「ethical AI」，但沒寫出對應控制，那不是治理，那是品牌文案。trustworthy、responsible、human-centered 這些詞都一樣，聽起來很順，落地卻常常是空的。\u003C\u002Fp>\u003Cp>我也整理過很多文件，裡面每段都在換同義字，像在比誰比較會寫字。通常那代表大家都不想寫出那句難看的話：規則是什麼、誰執行、失敗怎麼辦。\u003C\u002Fp>\u003Cp>實操寫法：把抽象標籤改成操作定義。\u003C\u002Fp>\u003Cul>\u003Cli>不要寫 transparent，改寫要揭露什麼。\u003C\u002Fli>\u003Cli>不要寫 fair，改寫要做哪種偏誤檢查。\u003C\u002Fli>\u003Cli>不要寫 safe，改寫哪個門檻會擋 release。\u003C\u002Fli>\u003Cli>不要寫 accountable，改寫誰核准、誰背書。\u003C\u002Fli>\u003C\u002Ful>\u003Cp>如果一個詞不能被翻成控制項，就直接刪掉。省事，也比較誠實。\u003C\u002Fp>\u003Ch2>節奏問題才是麻煩，政策得比模型跑得快\u003C\u002Fh2>\u003Cblockquote>“AI technology is rapidly evolving leading to a ‘pacing problem’ where traditional laws and regulations often cannot keep up.”\u003C\u002Fblockquote>\u003Cp>意思很簡單：你不能等法規完全講清楚才開始做控制。等規則定案，你的系統大概已經改兩輪了。真正該做的是，讓你的政策可以\u003Ca href=\"\u002Fnews\u002Fexpanding-flow-maps-variable-size-generation-zh\">跟著\u003C\u002Fa>變，不要每次都重寫。\u003C\u002Fp>\u003Cp>這點我很有感。很多團隊想等「最終答案」，但 AI 的現實不是這樣。你要做的是建立一個能吸收新規則的流程，而不是等一份完美政策。也就是說，風險審查、文件版本、變更管理、區域追蹤，這些都要先有。\u003C\u002Fp>\u003Cp>Wikipedia 也提到，硬法會慢，部分原因是 AI 應用太多樣，既有機關的管轄範圍又有限。這不是叫你躺平，是叫你把治理做成軟體：模組化、可\u003Ca href=\"\u002Fnews\u002Fopenai-test-model-broke-into-hugging-face-servers-zh\">測試\u003C\u002Fa>、可更新。\u003C\u002Fp>\u003Cp>實操寫法：做成活文件，不要做成 PDF 墳場。\u003C\u002Fp>\u003Cul>\u003Cli>高風險功能固定週期重審。\u003C\u002Fli>\u003Cli>政策版本跟 code 一樣留紀錄。\u003C\u002Fli>\u003Cli>按地區和產業追蹤法規變化。\u003C\u002Fli>\u003Cli>留一組可以快速加強的控制項。\u003C\u002Fli>\u003C\u002Ful>\u003Cp>政策一改，團隊要知道改了什麼、為什麼改。沒這件事，大家最後只會照舊做。\u003C\u002Fp>\u003Ch2>跨國協調很重要，因為 AI 本來就不守邊界\u003C\u002Fh2>\u003Cblockquote>“In 2023, the United Kingdom started a series of international summits on AI.”\u003C\u002Fblockquote>\u003Cp>白話翻譯就是：AI 規範不是單一國家的故事。頁面提到 \u003Ca href=\"\u002Ftag\u002Fai-safety\">AI Safety\u003C\u002Fa> Summit、AI Seoul Summit、AI Action Summit in Paris、AI Impact Summit in New Delhi。各國雖然細節不同，但都在試著對齊一些共同問題。\u003C\u002Fp>\u003Cp>這件事對做全球產品的人很重要。你的功能可能在一個市場上線，在另一個市場被審查，第三個市場的客戶又帶著不同期待來問你。如果你只看公司註冊地在哪裡，通常會漏掉真正的暴露面。\u003C\u002Fp>\u003Cp>我在 enterprise 案子裡也常看到這種狀況。某個區域的客戶要一份文件，另一個市場又要不同標準。最後你的「一份 policy」變成三份，然後大家還是覺得自己有統一治理。\u003C\u002Fp>\u003Cp>實操寫法：做 region-aware controls。\u003C\u002Fp>\u003Cul>\u003Cli>追蹤每個功能在哪些地區可用。\u003C\u002Fli>\u003Cli>存放各地需要的揭露文字。\u003C\u002Fli>\u003Cli>維護一張區域義務對照表。\u003C\u002Fli>\u003Cli>指定人追蹤跨境政策變化。\u003C\u002Fli>\u003C\u002Ful>\u003Cp>如果你賣全球，合規故事也得是全球版。只靠一份本地想像，撐不住。\u003C\u002Fp>\u003Ch2>可抄的模板\u003C\u002Fh2>\u003Cpre>\u003Ccode># AI Governance Policy Template\n\n## 1) Scope\nThis policy applies to all AI systems, models, prompts, datasets, and automated decision workflows used by the company.\n\n## 2) Definitions\n- AI system: any software that generates predictions, content, recommendations, classifications, or decisions using statistical or machine-learning methods.\n- High-impact use case: any AI feature that affects employment, credit, housing, healthcare, education, legal status, safety, access to services, or materially important decisions.\n- Human review: a trained person who can inspect, override, or block AI output before it is acted on.\n\n## 3) Ownership\nFor each AI feature, the team must name:\n- Business owner\n- Technical owner\n- Risk\u002Flegal reviewer for high-impact use cases\n- Incident responder\n\n## 4) Use-case inventory\nBefore launch, record:\n- Feature name\n- Purpose\n- Model\u002Fprovider\n- Data sources\n- User-facing outputs\n- Jurisdictions where it ships\n- Whether it is high-impact\n- Required disclosures\n\n## 5) Risk classification\nClassify each feature as:\n- Low risk\n- Medium risk\n- High risk\n\nA feature is high risk if it can influence rights, access, safety, or materially important decisions.\n\n## 6) Required controls\n### Low risk\n- Basic documentation\n- Logging enabled\n- User disclosure if AI output is visible\n\n### Medium risk\n- Documented testing\n- Human review for sensitive outputs\n- Bias and quality checks\n- Rollback plan\n\n### High risk\n- Formal approval before release\n- Written legal\u002Frisk review\n- Human override path\n- Audit log retention\n- Incident response procedure\n- Periodic re-review after launch\n\n## 7) Disclosure rules\nIf users interact with AI or rely on AI-generated outputs, the product must disclose:\n- That AI is being used\n- What the AI is used for\n- Any material limitations\n- Whether a human can review or override the output\n\n## 8) Testing and validation\nBefore release and after material changes:\n- Test for accuracy on representative cases\n- Test for bias and disparate outcomes where relevant\n- Test for prompt injection or misuse if the system is user-facing\n- Record results and remediation steps\n\n## 9) Change management\nAny material change to model, prompt, data source, threshold, or user workflow requires:\n- Versioning\n- Re-review of risk classification\n- Updated documentation\n- Approval for high-risk systems\n\n## 10) Incident response\nIf an AI system causes or may cause harm:\n- Pause or disable the feature if needed\n- Notify the owner and reviewer\n- Preserve logs and prompts\n- Investigate root cause\n- Record corrective actions\n- Decide whether customer notice is required\n\n## 11) Review cadence\n- Low risk: annual review\n- Medium risk: semiannual review\n- High risk: quarterly review and after any material incident\n\n## 12) Exceptions\nAny exception must be:\n- Written\n- Time-limited\n- Approved by the business owner and risk\u002Flegal reviewer\n- Reviewed again before expiry\n\n## 13) Attestation\nBy launching or operating this AI feature, the owners confirm that the required controls are in place and the documentation is current.\n\nOwner: ____________________\nDate: _____________________\nFeature: __________________\nRisk level: ________________\u003C\u002Fcode>\u003C\u002Fpre>\u003Cp>這份模板不花俏，但我就是要它不花俏。它把「我們應該做 AI policy」變成一份可運作的文件：有 owner、有風險、有審核、有事件處理。這樣才像真的能拿去跑。\u003C\u002Fp>\u003Cp>如果是我先落地，我會先做兩件事：先補 inventory，再補 ownership。這兩塊先補起來，很多混亂會直接少一半。接著加 disclosure 跟 incident response，因為那是最容易被拖到出事才補的部分。\u003C\u002Fp>\u003Cp>最後別把它丟在法務資料夾裡當裝飾。放到產品、工程、PM 真正在看的地方，讓它變成 launch gate 的一部分。沒人會看的 policy，本質上就是漂亮廢紙。\u003C\u002Fp>\u003Cp>原始拆解來源是 \u003Ca href=\"https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FRegulation_of_artificial_intelligence\">Wikipedia：Regulation of artificial intelligence\u003C\u002Fa>。我這篇的框架、白話翻譯和模板是我自己整理的，但底層概念、機構名稱與脈絡，主要衍生自該頁與其引用來源。\u003C\u002Fp>","把 AI 規範拆成可執行的控管清單，附可直接複製的政策模板，讓產品、工程、法務能一起用。","en.wikipedia.org","https:\u002F\u002Fen.wikipedia.org\u002Fwiki\u002FRegulation_of_artificial_intelligence",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784882006925-6jc3.png","industry","zh","e8475712-4e3e-47d0-9262-9057a52bdb1b",[17,18,19,20,21],"AI regulation","AI governance","risk classification","soft law","EU AI Act",[23,24,25],"AI 規範要用分層治理處理，不能只靠單一 policy。","把抽象原則翻成控制項、owner 與 review 流程，才算能落地。","可直接套用模板先做 use-case inventory、ownership 與 incident response。",1,"2026-07-24T08:32:55.521928+00:00","2026-07-24T08:32:55.485+00:00","21c4453f-222f-40cd-bc03-f930ac88600f",{"tags":31,"relatedLang":34,"relatedPosts":38},[32],{"name":18,"slug":33},"ai-governance",{"id":15,"slug":35,"title":36,"language":37},"ai-regulation-maps-the-rules-you-need-en","AI regulation maps the rules you need","en",[39,45,51,57,63,69],{"id":40,"slug":41,"title":42,"cover_image":43,"image_url":43,"created_at":44,"category":13},"61cd0573-fe03-4f6b-87cd-59ce185d37fe","sriram-krishnan-from-product-to-policy-zh","Sriram Krishnan 從產品走進政策","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784919780542-nd87.png","2026-07-24T19:02:35.366325+00:00",{"id":46,"slug":47,"title":48,"cover_image":49,"image_url":49,"created_at":50,"category":13},"bfb612ad-0fa0-4171-9128-118de64d0bfb","xai-july-2026-grok-45-updates-zh","xAI 7 月更新把 Grok 4.5 推進工作流","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784896372140-6lvy.png","2026-07-24T12:32:22.473907+00:00",{"id":52,"slug":53,"title":54,"cover_image":55,"image_url":55,"created_at":56,"category":13},"65f393a8-6d15-4046-a5f3-eece314a9a78","2026-2027-common-app-prompts-explained-zh","2026-2027 Common App 提示題解析","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784889180964-0ueb.png","2026-07-24T10:32:32.66725+00:00",{"id":58,"slug":59,"title":60,"cover_image":61,"image_url":61,"created_at":62,"category":13},"d363cbec-aa86-4767-ad50-42eb404536a6","agentic-ai-drew-four-regulators-week-zh","代理式 AI 一週內撞上四路監管","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784883784768-jrnc.png","2026-07-24T09:02:33.123514+00:00",{"id":64,"slug":65,"title":66,"cover_image":67,"image_url":67,"created_at":68,"category":13},"a3e9e018-6f39-4288-ab55-1b2a5536ad17","openai-status-july-2026-chatgpt-codex-zh","OpenAI 七月狀態頁很忙","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784809973710-s1rl.png","2026-07-23T12:32:29.665603+00:00",{"id":70,"slug":71,"title":72,"cover_image":73,"image_url":73,"created_at":74,"category":13},"6e295576-bf49-446d-85fe-55e16cf61885","amd-anthropic-2gw-mi450-ai-compute-deal-zh","AMD 與 Anthropic 的 2GW AI 算力合作","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1784808163551-3bx8.png","2026-07-23T12:02:20.252117+00:00",[76,81,86,91,96,101,106,111,116,121],{"id":77,"slug":78,"title":79,"created_at":80},"ee073da7-28b3-4752-a319-5a501459fb87","ai-in-2026-what-actually-matters-now-zh","2026 AI 真正重要的事","2026-03-26T07:09:12.008134+00:00",{"id":82,"slug":83,"title":84,"created_at":85},"83bd1795-8548-44c9-9a7e-de50a0923f71","trump-ai-framework-power-speech-state-preemption-zh","川普 AI 框架瞄準電力、言論與州權","2026-03-26T07:12:18.695466+00:00",{"id":87,"slug":88,"title":89,"created_at":90},"ea6be18b-c903-4e54-97b7-5f7447a612e0","nvidia-gtc-2026-big-ai-announcements-zh","NVIDIA GTC 2026 重點拆解","2026-03-26T07:14:26.62638+00:00",{"id":92,"slug":93,"title":94,"created_at":95},"4bcec76f-4c36-4daa-909f-54cd702f7c93","claude-users-spreading-out-and-getting-better-zh","Claude 用戶更分散，也更會用","2026-03-26T07:22:52.325888+00:00",{"id":97,"slug":98,"title":99,"created_at":100},"bd903b15-2473-4178-9789-b7557816e535","openclaw-raises-hard-question-for-ai-models-zh","OpenClaw 逼問 AI 模型價值","2026-03-26T07:24:54.707486+00:00",{"id":102,"slug":103,"title":104,"created_at":105},"eeac6b9e-ad9d-4831-8eec-8bba3f9bca6a","gap-google-gemini-checkout-fashion-search-zh","Gap 把結帳搬進 Gemini","2026-03-26T07:28:23.937768+00:00",{"id":107,"slug":108,"title":109,"created_at":110},"0740e53f-605d-4d57-8601-c10beb126f3c","google-pushes-gemini-transition-to-march-2026-zh","Google 把 Gemini 轉換延到 2026 年 3…","2026-03-26T07:30:12.825269+00:00",{"id":112,"slug":113,"title":114,"created_at":115},"e660d801-2421-4529-8fa9-86b82b066990","metas-llama-4-benchmark-scandal-gets-worse-zh","Meta Llama 4 分數風波又擴大","2026-03-26T07:34:21.156421+00:00",{"id":117,"slug":118,"title":119,"created_at":120},"183f9e7c-e143-40bb-a6d5-67ba84a3a8bc","accenture-mistral-ai-sovereign-enterprise-deal-zh","Accenture 攜手 Mistral AI 賣主權 AI","2026-03-26T07:38:14.818906+00:00",{"id":122,"slug":123,"title":124,"created_at":125},"191d9b1b-768a-478c-978c-dd7431a38149","mistral-ai-faces-its-hardest-year-yet-zh","Mistral AI 迎來最硬的一年","2026-03-26T07:40:23.716374+00:00"]