Anthropic Unveils Model Hardware Standard Alongside Surgical AI Milestones and Industry Moves

Anthropic Unveils Model Hardware Standard Alongside Surgical AI Milestones and Industry Moves The Rundown highlights Anthropic's entry into hardware control via its Model Hardware Standard, the UK's first live AI-assisted brain surgery, voice dictation workflows for directing agents, and various corporate updates across the AI landscape. Anthropic launched a research preview of the Model Hardware Standard to connect AI agents directly to physical lab and factory equipment. UK surgeons conducted the first live AI-assisted brain surgery to remove a pituitary tumor safely. Rowan highlights voice-whispering tools as an emerging paradigm for orchestrating AI agents. OpenAI, joined by 116 organizations, issued a warning regarding sophisticated AI-driven cyber threats. Reports indicate Nvidia plans to purchase Hugging Face for $12.9 billion, while Google launched Gemini Omni 1.1 Flash. Anthropic Model Hardware Standard Anthropic announced the Model Hardware Standard (MHS) in research preview, aiming to connect AI agents with factory machines, microscopes, and robotic arms using standardized natural language reference files instead of manual wiring. Setup times are expected to drop from weeks to hours or minutes. Tecan, QIAGEN, and AWS are partnering, with Hugging Face and Raspberry Pi adopting the standard ahead of a planned open-source release. AI in Live Neurosurgery Surgeons in the UK completed the first live AI-assisted brain surgery using a system developed by University College London. During the removal of an 11mm pituitary growth from patient Rhys Hibbert via his nose, the AI monitored the surgical camera feed to identify safe cutting zones and highlight vulnerable optic nerves and arteries. The patient's sight was restored within days, and expanded clinical trials are planned. Workflows and Directing AI by Voice Voice input apps like Wispr Flow are gaining traction as a fast, context-rich way to oversee multiple AI agents simultaneously. Additionally, the newsletter outlines Claude Code plugins like makerskills (/unstuck and /decide) to assist decision-making, and reader Karen D. shared a method using Google Gemini and whiteboard grids to catalog and evaluate family heirlooms. Industry Developments and Deals OpenAI published an open letter signed by 116 companies cautioning against advanced AI-enabled cyberattacks. Nvidia is reportedly acquiring Hugging Face for $12.9 billion, Google launched its Gemini Omni 1.1 Flash video model, Barret Zoph rejoined Google as a research VP for Gemini reinforcement learning, and Salesforce teamed up with Anthropic to release the Claudeforce plugin and integrate Claude as Slack's default model.

Anthropic发布模型硬件标准，英国开展首例实时AI辅助脑部手术 Anthropic推出旨在连接物理设备与实验室仪器的模型硬件标准研究预览版；同时英国医院完成首例实时AI辅助脑部手术，此外还涉及语音交互趋势及行业快讯。 Anthropic推出模型硬件标准（MHS），使AI代理能够操作显微镜、机械臂等工业与实验设备。 英国伦敦大学学院团队完成首例实时AI辅助脑外科手术，AI通过手术摄像头实时标记血管和视神经。 语音控制AI代理在开发者中兴起，被视为提高提示词上下文丰富度与多任务操作效率的方式。 OpenAI联合116家公司就AI网络攻击发出预警；英伟达据传拟以129亿美元收购Hugging Face；谷歌推出Gemini Omni 1.1 Flash；Salesforce与Anthropic合作推出Claudeforce。 Anthropic推出模型硬件标准（MHS） Anthropic发布了模型硬件标准研究预览版，允许AI代理识别、学习并运行工厂显微镜、机器人手臂等硬件，配置时间由数周缩短至几小时或几分钟。用户使用自然语言描述设备，MHS将其转换为参考文件供代理阅读。测试中Claude通过试错自主学会对准激光并编写了自动化脚本。该标准未来计划开源，合作伙伴包括Tecan、QIAGEN、AWS，Hugging Face和树莓派也将其整合进设备线。 英国完成首例实时AI辅助脑部手术 英国外科医生团队利用伦敦大学学院研发的AI系统，完成了首例实时AI辅助脑部手术。该系统在切除垂体肿瘤过程中实时分析手术摄像头画面，在独立监视器上标记安全切割点与需避开的血管及视神经。该AI基于数百个过往手术标注视频进行训练。患者术后数天内视力得到恢复，团队正准备开展更大规模试验。 语音低语控制AI成为新工作流趋势 旧金山开发者中正流行使用如Wispr Flow等语音工具控制AI代理。口述相较打字不仅速度更快（约每分钟150词对比60词），还促使提示词天然扩充2至3倍并包含更多上下文。配合苹果将在iOS中推出的AI听写功能，通过低语管理多个并行AI代理的工作流正在进一步普及。 行业快讯 OpenAI与Anthropic等116家机构联合签署公开信，警告AI驱动的网络攻击将更加复杂普及并呼吁协同防御；传英伟达将以129亿美元收购开源平台Hugging Face；谷歌发布升级版视频模型Gemini Omni 1.1 Flash，研究副总裁Barret Zoph重新加入谷歌负责Gemini强化学习；Salesforce与Anthropic合作推出包含37项销售技能的Claudeforce插件，并将Claude设为Slack默认模型。

^{Read Online} ^| ^{Sign Up} ^| ^Advertise **Good morning, {{ first\_name | AI enthusiasts }}, and welcome to our 2,879 new readers.** Anthropic’s Model Context Protocol (MCP) started as a connector and eventually became the default way AI agents talk to software. The company’s next act is getting a bit more physical. The new Model Hardware Standard plugs agents into the microscopes, robot arms, and assembly lines the world already runs, a simple layer that could suddenly make every machine AI-ready. **In today’s AI rundown:** Anthropic wires AI agents into real-world machines Rowan’s Corner: Whispering to AI agents Use these AI skills to make better decisions quickly UK trials live-video AI in brain surgery **LATEST DEVELOPMENTS** Image source: Anthropic **The Rundown:** Anthropic just [introduced](https://www.anthropic.com/news/model-hardware-standard-research-preview?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) the Model Hardware Standard in research preview, which lets AI agents look up, learn, and run factory machines like microscopes, robotic arms, and other lab gear that previously needed custom code — built to spread in an industry-standard way similar to the company’s Model Context Protocol (MCP). **The details:** Anthropic claims setup times of “hours or minutes” under the new spec, down from the weeks specialists currently spend hand-wiring instruments. Machine owners describe their equipment in natural language, with MHS turning it into a reference file agents can read to learn how to operate the device. In one test, Claude taught itself to align a laser by trial and error, then simplified the routine into a script that automated the job in a single pass. An open-source release is planned at a later date, with Tecan, QIAGEN, and AWS partnering and both Hugging Face and Raspberry Pi adding it to their device lines. **Why it matters:** Claude, for machines! Physical AI is the territory many companies are fighting for, and Anthropic just made it that much simpler to plug its frontier models into the microscopes, robot arms, and assembly lines those companies are chasing… In a way that won’t feel much different from connecting agents to software via MCP. **The Rundown:** Agents are non-deterministic and dependent on models, prompts, and tools that change without warning. Traditional engineering playbooks no longer translate. LangChain’s new guide breaks down the Agentic Operating Model, aligning people, process, and technology to build, deploy, and scale agents across your organization. **In the guide, you’ll learn:** Why AI agents don’t break like traditional software The engineering stack that covers the entire agent lifecycle How to shift from “build and deploy” to “operate and improve” Image source: *@tmiyatake1 on X* **Rowan:** Back in October of last year, I visited several startup offices in San Francisco and noticed the early innings of a trend — developers were using tools like Wispr Flow to multitask across workflows and control AI agents more efficiently, entirely by voice. Fast forward to today, and that trend has only accelerated. SF tech trends are fascinating because there’s something new every week, but most never escape the early-adopter bubble. Every so often, though, one breaks out and becomes huge for everyone. Now that Apple is shipping an AI dictation keyboard in iOS 27 (alongside a fully rebuilt Siri AI), I’m fully convinced: whispering to AI is breaking out of the bubble. I don’t take personal recommendations lightly, but this one's worth hopping on if you haven’t already. Here’s why: Speed and context. You speak at \~150 words per minute and type at *maybe* 60. But the underrated part is that spoken prompts are naturally 2–3x longer, so you include the context you’d skip when typing. It turns you into a director of agents. A common workflow is to have several tabs of agents open at once, focus on ONE window, whisper a 60-second brief, release, let the agent work, review, and move to the next. The AI dictation apps are so good that you can *literally* whisper to them, and it catches nearly everything. That’s the only way this works in an office of 50 people. If you’ve been following our Workflow Hub, you know one of my [favorite workflows](https://app.therundown.ai/community/posts/0d523183-b3ab-4b59-b341-ce757b16be1c?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) is whispering ideas into Wispr Flow on walks and between meetings, then using a scheduled Claude task to organize everything into my Notion overnight. Getting started is simple: either wait for the iOS drop, or skip the wait and map a Wispr-style app to your iPhone’s Action Button today. And if you want to go full Silicon Valley? Grab a $20 desk mic and direct your agents all day. **The Rundown:** In this guide, you will learn how to give your AI skills that make every decision easier. You’ll use an /unstuck tool, a /decide tool, and more to turn problems into solutions. **Step-by-step:** Open Claude Code in Terminal and install [makerskills](https://github.com/coreyhaines31/makerskills?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) with /plugin marketplace add coreyhaines31/makerskills, followed by /plugin install makerskills@makerskills. Use the repository’s routing table to match the problem to a skill: /unstuck when you keep circling the same wall or /decide when you need to choose between real options. Describe the situation and answer the follow-up questions. When social-fetch could not pull our X post and suggested a paid scraping API, /unstuck challenged the assumption that the video needed that demo. Try the framework on another job. /decide compared two hypothetical client projects, tested how reversible the choice was, recommended one, and set a date to revisit the decision. **Pro tip:** Run the skill inside a project or session that already has context on your work. You will get more specific questions and recommendations. **The Rundown:** Enterprises are quickly deploying and scaling AI agents, but at what cost? The latest Gartner webinar unpacks how leaders avoid the orchestration overlap and governance gaps that often undercut their efforts. **In this webinar, you’ll learn how to:** Cut through AI orchestration complexity Close critical governance gaps Build a trusted agentic AI framework Image source: UCLH / BBC **The Rundown:** UK surgeons just [announced](https://www.uclh.nhs.uk/news/first-patient-live-ai-assisted-sight-saving-brain-surgery?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) the first brain surgery ever performed with a live AI assistant, with a University College of London system watching in real-time through a surgical camera and flagging hidden arteries and optic nerves to avoid during the removal of a tumor. Patient Rhys Hibbert had no symptoms until a seizure in late 2024 revealed an 11mm pituitary growth that slowly narrowed his field of vision. Surgeons reached the tumor through Hibbert’s nose, while AI watched the camera feed on its own monitor and marked the safest spots to cut and areas to avoid. The AI trained on hundreds of labeled videos of past removals, with surgeon Hani Marcus saying it saw “more operations than most surgeons see or do in a lifetime.” The patient’s vision was restored within days following the successful removal, with the team now moving toward larger trials of the AI system. **Why it matters:** Robots may eventually be capable of the precise movements needed for most surgeries, but for now this is a powerful example of the near-term advances AI can bring to the operating table: an expert second set of eyes to monitor and flag while the human surgeon remains completely in control. **QUICK HITS** ▸ **Karen built a family heirloom catalog with Gemini and a whiteboard** Today’s workflow comes from reader [Karen D.](https://www.linkedin.com/in/karendevriescoaching?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925): “My mother recently moved into a retirement suite, which meant downsizing her possessions. We ended up with dozens of physical items she no longer had space for, and we wanted her children and grandchildren to have an opportunity to sign up for the family items they loved before anything was donated. I needed a way to catalog everything quickly, share the info remotely, and verify that we weren't accidentally giving away items of significant value. I tested a physical-to-digital workflow using Gemini. Instead of putting the items in a pile or typing out endless lists, I drew a numbered grid on a large whiteboard, placed one item in each square, and took a few photos. Gemini identified the objects by grid position, categorized them, estimated their market values, and built a ready-to-use spreadsheet with a signup column. The physical staging took about 30 to 45 minutes, and the entire digital workload took under 10 minutes. The process turned an overwhelming chore into something manageable.” *See Karen’s full workflow here. How do you use AI? Tell us for a chance to be featured.* 🔀 [Switch](https://flintai.dev/products/switch?utm_source=TheRundownAI&utm_medium=Newsletter&utm_campaign=NewTools082826) - Bring any AI agent into Slack, Teams, Discord, and more. Open source. No migration. No platform lock-in.\* 📄 [Cohere Parse 5](https://www.therundown.ai/tools/cohere-parse-5?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) - SOTA model for converting docs into machine-readable data ⚡️ [Gemini Omni 1.1 Flash](https://www.therundown.ai/tools/gemini-omni-1-1-flash?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) - Google’s upgraded video model 🎥 [H3 Max](https://www.therundown.ai/tools/h3-max?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) - fal's MiniMax H3 remixed video model tuned for speed and quality *\*Sponsored Listing* **OpenAI** [published](https://openai.com/collective-cyberdefense/?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) an open letter signed by 116 companies (including Anthropic) warning that AI-enabled cyberattacks will become “far more widespread and sophisticated” in the coming months, calling for a coordinated cyber defense push. **Nvidia** is reportedly [acquiring](https://www.theinformation.com/articles/nvidia-agrees-buy-open-source-model-repository-hugging-face-12-9-billion?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) open model hub Hugging Face for $12.9B, with open-source interest continuing to surge on Chinese model gains. **Google** [released](https://blog.google/innovation-and-ai/technology/developers-tools/build-with-gemini-omni-1-1-flash/?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) Gemini Omni 1.1 Flash, an upgraded video model that adds 40-sec scene extensions and 4K upscaling, moving to the top spot on Arena’s text-to-image leaderboard. **Thinking Machines co-founder Barret Zoph** is [rejoining](https://x.com/barret_zoph/status/2092790775524516206?s=20&utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) Google as a research VP to work on reinforcement learning for Gemini, coming after a brief second stint at OpenAI. **Salesforce** [partnered](https://www.salesforce.com/claudeforce/?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) with Anthropic to introduce Claudeforce, which puts a 37-skill sales plugin into Claude, while also making Claude the default model in Slack. Read our last AI newsletter: [The Ox Alpha mystery ends with Z.ai](https://www.therundown.ai/articles/the-ox-alpha-mystery-ends-with-z-ai?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) Read our last Tech newsletter: [Apple’s Mac Mini makes an AI comeback](https://tech.therundown.ai/p/apple-mac-mini-makes-ai-comeback?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) Read our last Robotics newsletter: [Cyborg roaches become tiny paramedics](https://robotnews.therundown.ai/p/cyborg-roaches-become-tiny-paramedics?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) Today’s AI tool guide: [Use these AI skills to make better decisions quickly](https://app.therundown.ai/guides/use-these-ai-skills-to-make-better-decisions-quickly?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) RSVP to next workshop on Sept. 2: [Build an AI operating system for your 1:1s](https://app.therundown.ai/live/build-an-ai-powered-operating-system-for-your-1-1s?utm_source=therundownai.beehiiv.com&utm_medium=newsletter&utm_campaign=rundown-0925) See you soon, *Rowan, Zach, Shubham, Jennifer, and Nate — the humans behind The Rundown*