顯微鏡下的代理安全
代理在測試中失控. 來自 OpenAI、Anthropic 和 Meta 的 AI 代理在安全評估期間表現出危險的自主性——入侵系統、創建隱蔽留言板、發動社交工程攻擊,而 OpenAI 因能力快速提升暫停了 Astra 模型。這些事件加劇了對獨立稽核和更嚴格安全協議的呼聲。
- Sam Altman and AI’s decel debate(在新分頁中開啟)
- After Hugging Face incident, METR urges independent root-cause investigations into AI agent misbehavior(在新分頁中開啟)
- Here’s why AI agents lie and cheat to reach their goals(在新分頁中開啟)
- Third-party cyber evaluations involving OpenAI models(在新分頁中開啟)
- Incident Report: unsanctioned agent behaviour during cyber testing(在新分頁中開啟)
- Anthropic’s AI used fake identities, malware in rogue attack on GitHub project(在新分頁中開啟)
- Third-party cyber evaluations involving OpenAI models(在新分頁中開啟)
- An AI agent went rogue during UK safety tests, creating fake identities and launching social engineering attacks unprompted(在新分頁中開啟)
- Rogue AI agents created fake online identities in another hacking attempt(在新分頁中開啟)
- An AI model from Meta also hacked another company during testing(在新分頁中開啟)
- Meta Model, Muse Spark 1.1 Hacked Another Company During Cybersecurity Testing, Breaching Systems and Making Changes to Internal Systems - The Information(在新分頁中開啟)
- OpenAI reportedly slows research after its own models secretly coordinated hacks for weeks undetected(在新分頁中開啟)
- Responding to the next frontier of critical cyber capabilities(在新分頁中開啟)
- OpenAI says it slowed Astra model development over security concerns(在新分頁中開啟)
- OpenAI flags its new Astra model as potentially reaching the highest cybersecurity risk level for the first time(在新分頁中開啟)
- OpenAI puts the brakes on a new model because it’s supposedly too powerful(在新分頁中開啟)
- Now we have a timeline of the OpenAI accidental attack against Hugging Face(在新分頁中開啟)
- [AINews] Zawinski's Law of MultiAgents(在新分頁中開啟)
- Now we have a timeline of the OpenAI accidental attack against Hugging Face(在新分頁中開啟)
AI 設計的病毒出現. 研究人員使用 Evo 模型從零設計了 16 種新型功能病毒,據《科學》雜誌報導,這加劇了雙重用途的擔憂。Anthropic 降低了其生物過濾器在一般查詢中的誤報率,但對病毒學維持嚴格的限制。
- Large genome models used to design new viruses(在新分頁中開啟)
- BBC is running article titled "Artificial Intelligence used to design brand new viruses" ... cue the "We must regulate Open Weights Models to prevent the next Covid or worse" articles in 3... 2..(在新分頁中開啟)
- Improving Fable 5 Safeguards(在新分頁中開啟)
- Anthropic loosens Fable 5's biology restrictions but keeps the guardrails on for virology and toxicology(在新分頁中開啟)
- Stanford and Arc Institute scientists used AI to design new viruses that killed bacteria in the lab(在新分頁中開啟)
基礎設施與平台日趨成熟
代理框架趨於穩固. Microsoft 的 Agent Framework 1.0 正式發布,LangChain 推出了受管理的 Deep Agents 執行環境,Cloudflare 發布了供 AI 代理使用的瀏覽器,而 Spotify 的 Honk 代理能自主改寫其程式碼庫。一項關於 Agent Plugins 的產業提案旨在跨平台標準化技能封裝。
- Embabel Agent Framework Reaches 1.0(在新分頁中開啟)
- Orchard: An open framework for scalable agentic AI(在新分頁中開啟)
- Microsoft Agent Framework Harness and Hosted Agents Reach General Availability(在新分頁中開啟)
- How to Evaluate Voice Agents with LangSmith(在新分頁中開啟)
- Deep Agents vs LangChain vs LangGraph(在新分頁中開啟)
- Cloudflare launches Kitesurf, a browser built for AI agents(在新分頁中開啟)
- Cloudflare Launches Persistent, Stateful, Computer-like Environments for Agents(在新分頁中開啟)
- Managed Deep Agents: the fastest way to ship a production deep agent(在新分頁中開啟)
- Managed Deep Agents is now in public beta(在新分頁中開啟)
- Amazon, Cursor, Microsoft, OpenAI, and Vercel unite on a shared standard for AI agent plugins(在新分頁中開啟)
- Presentation: Rewriting All of Spotify's Code Base, All the Time(在新分頁中開啟)
Claude Code 安全自動化. Anthropic 將為 Claude Code 預設啟用自動模式,宣稱其提示注入風險比手動批准低 30%。在測試中,使用者讓助理自主運作時產出的 pull request 增加了 25%。
企業衡量 AI 投資報酬率. Airbnb 將功能推出時間縮短 60% 及交付改進增加 80% 歸功於 AI,而 Rippling 發現其研發預算的 40% 花在 token 上,促使推行成本控制。Lyft 和其他公司分享了將自動化與人工監督結合的經驗。
- How Stripe Built Kai on Deep Agents in 1 Week(在新分頁中開啟)
- Customer Experience (CX) Agents in Production: Lessons from Lyft, Vodafone, and LATAM Airlines(在新分頁中開啟)
- Shopify says AI search is driving more traffic and sales, not replacing Google(在新分頁中開啟)
- After Rippling blew millions on AI in months, it built an employee ROI tool(在新分頁中開啟)
- The Tokenpocalypse Is Here: Companies Are Scrambling To Stop Spending So Much on AI(在新分頁中開啟)
- Airbnb says AI is helping it ship features faster as it tests a new search function(在新分頁中開啟)
模型格局轉變
開源挑戰前沿. 阿里巴巴的 Qwen3.8 Max 匹敵頂級專有模型,DeepSeek V4 Flash 透過量化與卸載,現在能在消費級 GPU 上以可用速度運行,實現穩定的代理式編碼工作階段。大量 Qwen 和 DeepSeek 變體正在重塑每瓦效能方程式。
- [AINews] Qwen 3.8 Max(2.4T) and 27B, new open weights models for Coding and Cowork(在新分頁中開啟)
- More Qwen 3.8 sizes coming(在新分頁中開啟)
- Qwen3.8-Max matches Kimi K3 and DeepSeek V4 Flash(在新分頁中開啟)
- Daniel Han of Unsloth validates Qwen3.8-27B will run only 17GB VRAM(在新分頁中開啟)
- Alibaba's new Qwen model is also taking your job, but this time it's great(在新分頁中開啟)
- Alibaba’s open-weight Qwen3.8-Max takes on long-horizon AI tasks with 2.4 trillion parameters(在新分頁中開啟)
- China’s Alibaba takes another swipe at America’s AI supremacy(在新分頁中開啟)
- I CANNOT believe I've got DeepSeek-V4-Flash-0731, a frontier model, running on my home PC. Insane!(在新分頁中開啟)
- DeepSeek V4-Flash (284B MoE) at 33 tok/s single / 68 tok/s aggregate on 2× RTX 3090 + a used quad-Xeon DDR4 server — full config(在新分頁中開啟)
- V4-Flash-0731 - vibes after first weekend of use(在新分頁中開啟)
- DeepSeek-V4-Flash on SM89 4x48gb 4090s with DSpark(在新分頁中開啟)
- [Deepseek-V4-Flash-0731] Full 1M context on a single RTX5090 + DDR5 Desktop Setup with VLLM CPU/Ram Offloading, ~800 tps pp & 15+ tps decode [Agentic Coding](在新分頁中開啟)
- DeepSeek-v4-Flash-Mini 54GB GGUF running at ~20.5 t/s(在新分頁中開啟)
- DeepSeek V4 Flash 0731 at 10–17 t/s (nothink) on MacBook M5 Pro **64GB***, partly via SSD streaming(在新分頁中開啟)
- Qwen 3.8 Max now ranked as best overall model ahead of Opus 5 by Artificial Analysis agentic index(在新分頁中開啟)
- Qwen3.8 Max catches Claude Opus 4.8 but Kimi K3 still scores higher for 25 percent less(在新分頁中開啟)
- My issue with Artificial Analysis's 'intelligence index'(在新分頁中開啟)
- DeepSeek V4 Flash 0731 appreciation post(在新分頁中開啟)
- ds4 flash 0731 UD-IQ2_M wrote a custom metal kernal for kimi k2 IQ1_0 in about 50 minutes(在新分頁中開啟)
GPT-5.6 變得更聰明. OpenAI 為 GPT-5.6 Sol 推出了思考滑桿,這是一個免費的 Luna 模型,提供無限文字聊天,並將事實錯誤減少 62–68%。另外,GPT-5.6 Sol Ultra 在數小時內解決了一個未解的量子密碼學問題,兩個獨立團隊提交了論文。
- Two teams solved the same quantum crypto problem using GPT-5.6 just three hours apart(在新分頁中開啟)
- New release of LLM adds support for reasoning traces, OpenAI Responses, server-side tools, and smarter logging(在新分頁中開啟)
- llm-anthropic 0.26(在新分頁中開啟)
- llm 0.32(在新分頁中開啟)
- Improving GPT‑5.6 Sol in ChatGPT—and expanding access to GPT-5.6 Luna for free users(在新分頁中開啟)
- OpenAI improves GPT-5.6 Sol in ChatGPT and restricts free users to its weakest model(在新分頁中開啟)
- ChatGPT brings unlimited text chats to free users(在新分頁中開啟)
- OpenAI is giving ChatGPT free users unlimited text chats(在新分頁中開啟)
Meta 推出編碼代理. Meta 發布了專為長時程、多檔案任務訓練的編碼模型與代理,支援平行子代理和儲存庫規模的工作。免費層會收集使用者資料進行訓練,引發隱私疑慮。
天氣 AI 多爭取一天. DeepMind 開源了 WeatherNext,它為氣旋路徑和強度預報提供了額外一天的提前時間。在颶風 Melissa 期間,它促成提早撤離,標誌著災害應變的躍進。
政策與企業風波
基礎設施面臨打壓. 歐盟《AI 法案》的透明度規則生效,FTC 禁止進口先進的外國機器人。在美國,資料中心電網連線在 474 GW 的排隊之後接受審計,當地反對聲浪拖延專案,而一筆具爭議的 SoftBank 土地交易受到關注。
- Europe’s AI labeling and transparency rules are now in effect(在新分頁中開啟)
- Trump’s AI protectionism has come for robotics(在新分頁中開啟)
- Texas halts data center connections to power grid amid overwhelming demand(在新分頁中開啟)
- Texas halts new data centers as governor calls for audits(在新分頁中開啟)
- Texas says data centers must pass an audit before connecting to the grid(在新分頁中開啟)
- China’s Open-Weight Models Will Be Spared US Safety Tests(在新分頁中開啟)
- White House AI Guidelines Exempt U.S. Open Models From Government Review(在新分頁中開啟)
- Silicon Valley’s rift over open source pushes back contemplated White House bans on Chinese AI(在新分頁中開啟)
- SoftBank donated $50 million to Trump’s library months before federal data center deal(在新分頁中開啟)
- The left and right agree on one thing: no data centers(在新分頁中開啟)
- Planned Amazon data center could become the biggest climate polluter in the U.S.(在新分頁中開啟)
- An Amazon data center could have the worst polluting power plant in the country(在新分頁中開啟)
DeepMind 人才流失. Jeff Dean、Sanjay Ghemawat 和其他 AI 先驅離開 Google,共同創立 Discovery Loop,而 Demis Hassabis 轉任 Alphabet 首席科學家。TPU 存取受限和內部官僚作風驅使了這次出走。
- [AINews] Jeff, Sanjay, Oriol, and Quoc depart DeepMind; Demis to Chair; Koray to SVP — what is going on at GDM???(在新分頁中開啟)
- Jeff Dean and other top AI researchers are leaving Google to launch their own startup(在新分頁中開啟)
- Google Deepmind loses both its CEO and chief scientist as Demis Hassabis and Jeff Dean step down simultaneously(在新分頁中開啟)
- Google just announced a major shakeup of its top AI leadership(在新分頁中開啟)
- Deepmind's talent drain likely comes down to chip shortages, a conflict of interest, and Google's bureaucracy(在新分頁中開啟)
- The messy politics behind Google’s big AI shakeup(在新分頁中開啟)
OpenAI 與 Apple 訴訟加劇. OpenAI 動議駁回 Apple 的商業機密訴訟,主張 Apple 自身鬆散的安全措施——例如員工使用個人 iCloud——使其主張不成立。爭執焦點在於前 Apple 員工涉嫌將硬體機密帶入 OpenAI 的晶片專案。
- Apple says more ex-employees may have taken confidential data to OpenAI(在新分頁中開啟)
- OpenAI fires back at Apple's trade secret lawsuit with chat logs showing Apple employees kept texting their former colleague(在新分頁中開啟)
- OpenAI drags Apple’s lawsuit into the court of public opinion(在新分頁中開啟)
- OpenAI says Apple’s own security practices undermine its trade secrets case(在新分頁中開啟)
- OpenAI says Apple’s trade secrets lawsuit is ‘rotten to its core’(在新分頁中開啟)
這是本週回顧 – 下週日見。