Nvidia CEO Jensen Huang Calls AI Cybersecurity Panic a Sales Tactic ↗
Nvidia CEO Jensen Huang said the panic over AI cybersecurity is hysteria manufactured by the industry to sell products. He argued AI security risks are overblown and are essentially a sales strategy by security vendors. The view aligns with the Replit CEO's comment from the previous day.
Editor’s take
Calling all AI security panic a sales pitch is like mistaking every fire alarm for a sales bell. The trouble is that a sales bell and a fire alarm can look identical. Human judgment does not need to be led by the claims of vendors or leaders.
Read the evidence
- Original source
- Nvidia CEO Jensen Huang 称 AI 网络安全恐慌是"行业为推产品制造的歇斯底里"——@CoinMarketCap:Huang 直言 AI 安全风险被过度渲染,本质是安全厂商的销售策略。与 Replit CEO 前日观点一致。
Read the sourceChatGPT top tier subscription discontinued; lower tiers remain as OpenAI adjusts pricing ↗
The top ChatGPT subscription tier has been discontinued, while lower tiers remain available. OpenAI is adjusting its pricing levels.
Editor’s take
The end of the top tier looks like an adjustment on the way from AI as a privilege for some toward something like a utility bill. Pricing changes decide less about features than about who benefits, and how quickly. In a convenient daily life, a changing price tag becomes quiet sorting.
Read the evidence
- Original source
- ChatGPT 20x(最高档)官方停止订阅——@yanhua1010:5x 和 Plus 仍可用。Astra 不开 Max 的话 5x 日常足够。OpenAI 在调整定价层级。
Read the sourceGrok Bot open-sources six bots, hits 1M+ views — ecosystem becomes social engine ↗
Several Grok Bot bots have been open-sourced and drawn massive views. The Grok Bot ecosystem is moving from a mere tool toward a social propagation engine.
Editor’s take
AI that generates its own momentum is like ripples spreading from a stone tossed into a pond, amplifying beyond direct human effort. This marks a shift in how ideas circulate, while making it harder to tell what is organic.
Read the evidence
- Who it affects
- Young readers interested in AI, startups, technology, and how information and bots spread on social platforms.
- Original source
- Grok Bot 开源 6 个 bot 带来 100 万+浏览量——@jasonzhou1993:已开源至 treg.to/grokbot#bots。Grok Bot 生态从工具走向社交传播引擎。
Read the sourceGPT-6 Astra Pro offered free to verified US clinicians ↗
An OpenAI figure announced free access to a new model for verified US clinicians through ChatGPT for Clinicians. Individual eligibility and broader applicability remain unclear.
Editor’s take
A free clinical tool that supports judgment could narrow gaps in knowledge. Like the stethoscope becoming standard equipment, it may alter the rituals of the exam room itself. Yet who is allowed to use it will shape the quality of care.
Read the evidence
- Who it affects
- Readers interested in AI in medicine, professional workflows, and the regulation of clinical tools.
- Original source
- GPT-6 Astra Pro 向临床医生免费开放——@gdb(OpenAI):经验证的美国临床医生今日起可通过 ChatGPT for Clinicians 免费使用 GPT-6 Astra Pro。AI 进入医疗专业领域的里程碑。
Read the sourceCrowdsourced Challenge Cuts Estimated Quantum Attack Cost on Bitcoin and Ethereum ↗
A crowdsourced challenge pushes the estimated cost of quantum attacks on Bitcoin and Ethereum below an earlier benchmark, narrowing the window for migration.
Editor’s take
Every time the estimated cost of breaking a system falls, the ground under long-held assets feels thinner. Clinging harder only builds a wall between a person and the world, and the anxiety keeps feeding itself. Faced with a bridge that already creaks, choosing not to cross it can feel lighter than crossing.
Read the evidence
- Who it affects
- A reader holding crypto or stocks who watches how technology shifts affect savings and cross-border life.
- What remains unknown
- The assumptions behind the estimate, the scope of the challenge and who verified it need checking at the source.
- Original source
- 量子计算攻击 BTC/ETH 成本估算被众包挑战降低 50%+——@laurashin:一项众包挑战将攻击比特币和以太坊的量子计算成本估算压低至 Google 三月基准的一半以下。加密货币量子安全时间表再收紧。
Read the sourceSEC proposes blockchain ledger as official record of securities ownership ↗
SEC proposes blockchain ledgers as official records of securities ownership; a commentator says this could cut dual bookkeeping and help asset regulation.
Editor’s take
If ledgers become official records, ownership moves from paper to a computational layer. Reducing dual bookkeeping is less like a lighter wallet and more like ending a double life of accounts. Regulatory recognition is weighty because it draws the boundary where technology touches public documents.
Read the evidence
- Original source
- SEC 提议区块链账本可作为证券所有权正式记录——@therealchaseeb:可减少代币化证券当前的双重记账要求,RWA 基础设施迈出监管认可关键一步。
Read the sourceCLARITY Act revision adds DeFi registration rule ↗
Senate Republicans released a revised CLARITY Act adding a requirement that DeFi protocols not truly decentralized register with the CFTC. A cloture vote is scheduled.
Editor’s take
The revision pulls the decentralization story back into the human world: registration and disclosure work like traffic lights at a crossing, not to stop someone but to let strangers gauge risk. The real question is less whether the slogan stays pure than whether cross-border lives get a predictable lane. A market held up by belief alone resembles driving a night road without guardrails, where distance turns more and more into luck.
Read the evidence
- Original source
- CLARITY Act 修订版发布+DeFi 新规——@laurashin:参议院共和党发布修订版,新增"非真正去中心化的 DeFi 协议必须向 CFTC 注册"要求。终止辩论投票定于周二 2:15pm ET。
Read the sourceAINEW
PRIMARYGitHub Releasescloudflare/workers-sdk create-cloudflare@2.72.7 ↗
11 Sept, 23:22 JST
Newly created projects default to compatibility date 2026-09-11, tracking pinned workerd 1.20260911.1; C3 ships this to npm create cloudflare users.
Editor’s take
A quiet change to a default is like arriving at a new apartment and finding the building rules already rewritten before you get the key. Tools that move with a runtime set the assumptions of a world before makers have fully imagined it. That unglamorous update shapes the silhouette of services yet to be built.
Read the evidence
- Original source
- ### Patch Changes - [#15604](https://github.com/cloudflare/workers-sdk/pull/15604) [`9109a76`](https://github.com/cloudflare/workers-sdk/commit/9109a767a47defa3e060632b3e39f918ed628c27) Thanks [@dom96](https://github.com/dom96)! - Update the default compatibility date for newly created projects to 2026-09-11 The default compatibility date tracks the pinned `workerd` release, which was bumped to `1.20260911.1`. Releasing C3 ships the updated date to `npm create cloudflare` users, since it is bundled into the published package.
Read the sourceAINEW
PRIMARYGitHub Releasescloudflare/workers-sdk wrangler@4.131.1 ↗
11 Sept, 23:22 JST
Wrangler adds a note that DNS and TLS provisioning may continue after a deploy adds a custom domain or enables Preview URLs; stable redeploys do not repeat it.
Editor’s take
A tool that explains its own process is a sign of how much of the work machines have quietly taken over. Like a notice board at the edge of a construction site, it tells anyone passing by that the job is still in progress.
Read the evidence
- Original source
- ### Patch Changes - [#15592](https://github.com/cloudflare/workers-sdk/pull/15592) [`945aaa3`](https://github.com/cloudflare/workers-sdk/commit/945aaa32e0c38116e501f8509884cb1cc8f1d51b) Thanks [@WillTaylorDev](https://github.com/WillTaylorDev)! - Add a provisioning delay note when custom domain Preview URLs change Wrangler now explains that DNS and TLS certificate provisioning may continue after a deploy adds a custom domain or enables its Preview URLs. Stable redeploys don't repeat the note. This assumes that a request which matches the stored custom domain state doesn't restart provisioning. The client infers this from the API changeset and current domain record because this repository can't verify the backend behavior. - [#15592](https://github.com/cloudflare/workers-sdk/pull/15592) [`945aaa3`](https://github.com/cloudflare/workers-sdk/commit/945aaa32e0c38116e501f8509884cb1cc8f1d51b) Thanks [@WillTaylorDev](https://github.com/WillTaylorDev)! - Clarify production status labels for custom domain routes Wrangler now prefixes explicit custom domain production states with `production:` so they match Preview labels. The updated labels appear in deployed trigger output and `WRANGLER_OUTPUT_FILE_PATH`. - [#15602](https://github.com/cloudflare/workers-sdk/pull/15602) [`47d906f`](https://github.com/cloudflare/workers-sdk/commit/47d906f52d109509f61b1c801c1b08ecad583c0d) Thanks [@dependabot](https://github.com/apps/dependabot)! - Update dependencies of "miniflare", "wrangler" The following dependency versions have been updated: | Dependency | From | To | | ------------------------- | ------------- | ------------- | | @cloudflare/workers-types | ^5.20260910.1 | ^5.20260911.1 | | workerd | 1.20260910.1 | 1.20260911.1 | - [#15592](https://github.com/cloudflare/workers-sdk/pull/1559
Read the sourceCursor launches Projects mode for AI coding with a persistent coordinator agent ↗
Cursor introduced Projects mode, replacing per-task chats with a coordinator agent in a persistent thread. AI coding tools shift from conversations to projects.
Editor’s take
Closing a conversation after each task is like changing shop counters for every errand. A persistent thread with a coordinating agent keeps a shared workbench open and reduces the loss of context. As AI coding shifts from chats to projects, the human role moves from doing each step toward editing intent.
Read the evidence
- Original source
- Cursor 发布 Projects 模式——@cursor_ai:告别每个任务一个 chat,改为与 coordinator agent 在单一持久线程中工作。类似 @bot 的持续对话范式,AI 编码工具架构正在从"对话"走向"项目"。
Read the sourceChatGPT for Financial Services Launches for Investment Banking and Equity Research ↗
OpenAI has launched a ChatGPT workspace for investment banking and equity research. It integrates financial data sources such as Daloopa, PitchBook, LSEG, Crunchbase, and Quartr. Analysts’ daily workflows are being taken over by AI, and many financial services apps face replacement.
Editor’s take
When financial data flows into a conversational surface, the center of analysis shifts from gathering information to judgment and asking better questions. It is like packing an entire financial data assembly line into a chat window. From an AI-centered view, apps that merely move data around will quickly lose their presence. The human role narrows to designing questions and reading context.
Read the evidence
- Original source
- ChatGPT for Financial Services 上线——@oragnes:OpenAI 直接面向投行和股票研究推出工作台,集成 Daloopa/PitchBook/LSEG/Crunchbase/Quartr 等金融数据源。分析师日常工作流被 AI 接管,大量金融服务应用面临替代。
Read the sourceFable data leak deepens: researcher claims purchase from LLM reseller ↗
Researcher Chaofan Shou claims a Fable dataset from a Chinese LLM reseller held corporate SSH keys, VPN configs and cloud tokens, raising supply-chain concerns.
Editor’s take
The episode is like a letter forwarded many times, its original sender hard to trace by the time it is opened. It shows convenience resting on outsourced trust, where supply-chain edges matter more than any single intruder. The more data gathers, the more those edges blur.
Read the evidence
- What remains unknown
- The claim is not independently verified; scope, affected organizations and legal impact remain unclear.
- Original source
- Fable 数据泄露事件持续发酵——@MaxForAI:安全研究员 Chaofan Shou 声称从中国头部 LLM 中转商处购得 6TB Fable 数据,含企业 SSH Key/VPN 配置/阿里云 Key/GitLab Token 等,足以入侵 7 家政府机构+多家科技公司。数据供应链安全问题升级。
Read the sourceOpenAI to host its agent harness and sandbox as a free cloud service ↗
The harness and sandbox behind Codex are being packaged as a free hosted service, with sandboxes selectable from several vendors. The race moves from model access to agent runtime.
Editor’s take
OpenAI is packaging the harness and sandbox behind Codex as a free hosted service, with sandboxes selectable from several vendors — a real gain in freedom for independent developers and small teams, who no longer need their own infrastructure to run agents. Picture it as lending out the whole kitchen for free rather than just selling recipes and ingredients. Yet the keys to the runtime stay in a few cloud vendors' hands, and a free kitchen can be reclaimed whenever the rules change. It moves toward creative freedom while quietly handing the rope of dependency to someone else.
Read the evidence
- Who it affects
- Developers and teams choosing an agent stack, including solo builders in Japan who now have more runtime options.
- Original source
- OpenAI Agents API 行业影响——@aigclink:将驱动 Codex 的 harness+sandbox 打包成免费云托管服务,行业从 token 工厂竞争转向 Agent Runtime 竞争。沙箱可选 OpenAI/Blaxel/Cloudflare/E2B/Modal 等。
Read the sourceSpaceXAI to livestream a company built with Grok Bot alone ↗
A three-person team will build a business from scratch on camera using Grok Bot over a short run of days. It is a public test of the AI-native company idea.
Editor’s take
Turning a small team building a company on camera with an AI into a livestream is a performance, not proof about whether AI can replace people. If the tool genuinely lowers the barrier so that fewer hands can turn an idea into a living, that is an expansion of human capability and deserves credit. It is like assembling a banquet from instant seasoning packets: edible, but you cannot see the cooking. What decides the meaning is who benefits from the labor saved and who carries the work displaced.
Read the evidence
- Who it affects
- Founders and small teams curious how far AI tooling carries a business, and sceptics watching for the limits.
- What remains unknown
- The format, roles and how success will be judged are not described in the source.
- Original source
- SpaceXAI 将直播三人团队用 Grok Bot 从零构建公司——@elonmusk:9/15-17 三天直播,mattyp/poteto/roshan_s 全程用 Grok Bot 创业。AI 原生公司从概念实验走向公开直播挑战。
Read the sourceMiro said to be acquired by Bending Spoons as collaboration SaaS reshuffles ↗
The whiteboard vendor is reportedly being bought by Bending Spoons at a modest multiple of revenue, read as another sign of AI pressure on collaboration software.
Editor’s take
A collaboration tool being bought at a modest price by a private-equity style owner is the kind of reshuffle where book gains go to a few shareholders while the risk lands on employees and users, showing up as higher renewal prices, thinner service and cut teams. It is like a favorite restaurant sold to a chain: the sign stays, the menu and the portions change. Buying and selling capital is not a crime in itself, but treating people's jobs and dependencies as assets to be shuffled at will is where the harm lies.
Read the evidence
- Who it affects
- SaaS watchers and teams in Japan who run Miro daily and depend on its terms.
- What remains unknown
- Deal value, timing and effects on existing contracts or pricing are unverified.
- Original source
- Miro $17.5B→被 Bending Spoons 收购(~2.3x ARR)——@deedydas:继 Airtable 之后又一 SaaS 巨头被低估收购,协作工具赛道被 AI 冲击的缩影。
Read the sourceModel rebuilds aircraft landing gear mechanism from video alone ↗
A post describes a model understanding and reconstructing a retractable landing gear mechanism from a video, which earlier models could not reproduce.
Editor’s take
A model rebuilt the retraction and extension of an aircraft's landing gear from a single video — machines moving from recognising to assembling — and knowledge reaching more hands is a good thing. It is like a disassembly exercise once taught only to the top students being opened up in front of the whole class: lower the threshold and more people can work with their hands. What deserves watching is who uses that power, and whether it repairs things or aims at people; the tool itself neither saves nor kills.
Read the evidence
- Who it affects
- Developers and researchers evaluating multimodal models
- What remains unknown
- The result has not been independently verified.
- Original source
- GPT-6 Astra 机械结构逆向工程——@FinanceYF5:仅凭一段 YouTube 视频,Astra 理解并重建了 Cessna 337 Skymaster 起落架的完整收放机制,其他模型此前一直无法正确还原。多模态理解从"看懂"到"建模"的跨越。
Read the sourceGitHub invites Copilot users to a build campaign ↗
The campaign asks people to create with the Copilot app and command line tool, with shop credit offered to participants.
Editor’s take
A campaign that lures people to build with points and store credit is no grand narrative, but laying tools out for anyone to try and change carries the smell of openness. It is like a neighbourhood canteen holding a free tasting day: the first bite costs nothing and whether you cook is your own call. It oppresses nobody and frees nobody; it just makes the ticket to making things a little cheaper — freedom often looks this unglamorous.
Read the evidence
- Who it affects
- Developers and makers who use Copilot and can meet the campaign terms.
- What remains unknown
- Individual eligibility and regional limits are not stated and must be verified.
- Original source
- @github - GitHub Copilot Day:Copilot app/CLI 创作活动,$100 GitHub Shop credit 奖励。
Read the sourceAI safety forum asks how far trust can go ↗
A Stanford founders group and an AI organisation are hosting a security discussion, with a foundation keynote planned.
Editor’s take
A forum that puts 'can we trust AI we do not fully understand' on the public table beats deciding behind a closed door. It is like someone in the school building insisting on checking the fuse box: the check may find nothing, but at least the lights are on while people look. Dignity means having the right to know what system decides for you, rather than being told to just use it with confidence; a place that asks such questions deserves a little of our time.
Read the evidence
- Who it affects
- Researchers, developers and students following AI safety and governance.
- What remains unknown
- Eligibility, whether it is streamed and the full agenda are unconfirmed.
- Original source
- @0G_labs - "Can we trust AI we don't fully understand?"——Stanford Founders Club 联合 House of AI 安全论坛,Linux Foundation 主旨演讲。
Read the sourceUpdated US crypto market-structure bill moves to next stage ↗
Senators released a revised version with DeFi requirements and credit union amendments, ahead of a procedural vote.
Editor’s take
Writing rules for a sector thick with scams and leverage treats ordinary people's money as something that matters, which is closer to dignity than letting the free-for-all run. But the text is drafted by people who know the industry well, and what the DeFi clauses cover or wave through decides whether this is a guardrail or a signboard. It is a school canteen menu: supervision beats none, yet if the supplier writes the menu, the children eat from the same pot. What deserves praise is transparency and accountability, not deregulation with a new label.
Read the evidence
- Who it affects
- Crypto firms and investors tracking US regulation
- What remains unknown
- Passage is not assured and the scope may still change.
- Original source
- CLARITY Act 最新版发布——@a16zcrypto:参议院发布含 DeFi 要求和信用合作社修正的更新版 CLARITY Act,下一轮推进投票定于 9 月 15 日周二 2:15pm ET。美国 crypto 监管框架进入关键立法阶段。
Read the sourceCherry Studio mobile adds GitHub support ↗
Cherry Studio mobile can now connect to GitHub to manage repositories, expanding mobile AI development capabilities.
Editor’s take
Letting a phone connect to code repositories and manage projects puts a tool of creation back into a pocket rather than keeping it on an office desk, and that widens who can take part in building things. It is moving the kitchen from a big canteen to your own stove: the cooking may or may not be good, but at least the doorman is out of the way. Nothing here is dramatic and nobody is harmed; the value is a threshold pressed a little lower. The verdict is positive — the more people who can build for themselves, the less concentrated power becomes.
Read the evidence
- Who it affects
- Developers using Cherry Studio
- Original source
- Cherry Studio 移动端支持插件——@rwayne:Cherry Studio 移动端现可连接 GitHub 管理仓库,AI 开发工具的移动端能力持续扩展。
Read the sourceOpenAI's Navier-Stokes release included a Lean 4 formal proof ↗
11 Sept, 06:22 JST
The OpenAI Navier-Stokes release included a formal proof.
Editor’s take
A release like this is like adding a checkable map to a mathematical labyrinth. If it is consumed only as a slogan, it remains a polished signboard; if it enters public verification, it can become part of accumulated knowledge.
Read the evidence
- Original source
- OpenAI’s Navier-Stokes release included a Lean 4 formal proof
Read the sourceCognition launches new SWE-2 model, rivaling Fable 5.1 and GPT-Astra ↗
11 Sept, 00:29 JST
Cognition launched a new model and says it competes with several leading models.
Editor’s take
Model competition is like the readout on a treadmill: the display changes, but that does not prove the daily run has become easier. The meaningful test is whether ordinary coding work actually changes for the better.
Read the evidence
- Original source
- Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra
Read the sourceAINEW
PRIMARYGitHub Releaseshonojs/hono fixes XSS in JSX boundary components ↗
5 Sept, 04:18 JST
GitHub release: hono/jsx boundary components such as Suspense, ErrorBoundary and Context.Provider left plain strings unescaped, causing XSS.
Editor’s take
Code boundaries are like the window sashes of a house. Just as outside air slips through unseen gaps, untrusted strings can cross boundaries and mix into what is displayed. The fix shows a reality: convenience rests on the accumulation of small escapes.
Read the evidence
- Original source
- ## Security fixes This release includes a fix for the following security issue: ### `hono/jsx` renders plain strings unescaped in boundary components, leading to XSS Affects: `Suspense`, `ErrorBoundary`, and `Context.Provider` in `hono/jsx`, and `renderToString()` / `renderToReadableStream()` in `hono/jsx/dom/server`. Fixes missing HTML escaping for a plain string placed directly as a child or `fallback` of these components, or as the root value of the server rendering functions, so untrusted strings could be emitted as markup. GHSA-hxh3-vqpv-xpqv --- Users who render untrusted strings inside `Suspense`, `ErrorBoundary`, or `Context.Provider`, or pass them directly to `hono/jsx/dom/server`, are strongly encouraged to upgrade to this version.
Read the source