SunBrief#95: Anthropic Moves AI Agents Into the Physical World

Welcome to the SunBrief

Today in SunBrief 🌞

• Supercharge Your Productivity With 1,000+ ChatGPT Prompts

• Anthropic Introduces Hardware Standard for AI Agents

• Stock Updates

• Google Gives Gemini Omni 1.1 Flash More Control Over AI Video

• Put Your Own AI Agent to Work

• AI Helps Surgeons Remove a Brain Tumour in a World First

• AI Highlights of the Week

• Too Important to Miss

Supercharge Your Productivity With 1,000+ ChatGPT Prompts from Superhuman AI

ChatGPT is insanely powerful. But most people waste 90% of its potential by using it like Google. Master AI and work 10X faster with 1,000+ proven ChatGPT prompts. Sign up for Superhuman AI to get:

• 1,000+ ready-to-use prompts to solve problems in minutes instead of hours—tested & used by 1M+ professionals

• Superhuman AI newsletter (3 min daily) so you keep learning new AI tools & tutorials to stay ahead in your career—the prompts are just the beginning

Anthropic Introduces Hardware Standard for AI Agents

The Model Hardware Standard connects lab instruments and factory machines through one shared interface

Anthropic has opened a research preview of the Model Hardware Standard, or MHS, developed with HHMI Janelia. It gives AI agents a common way to understand and operate programmable devices such as microscopes, liquid handlers, robotic arms, and quantum computing equipment.

Key Points:

• One Language for Hardware: MHS translates different machines into simple commands, removing the need for a custom connection for every device.

• Setup Gets Much Faster: Anthropic says labs can connect equipment in hours or minutes instead of waiting weeks or months.

• Agents Can Run Experiments: AI can coordinate several instruments, adjust settings as results change, and sometimes recover from hardware errors.

• Early Tests Look Promising: QuEra used MHS to restore a quantum computer laser lock 99.3% of the time without human help.

• It Is Built to Be Open: MHS works with programmable devices and different AI models through MCP, command lines, and APIs.

• Human Oversight Still Matters: The preview remains limited because current models can misunderstand physical failures and cannot control devices without programmable interfaces.

Why It Matters:
MHS could do for physical equipment what common software standards did for apps: make different systems easier to connect and automate. If it works safely, labs and factories could run more experiments around the clock, but errors now carry physical consequences, making expert oversight essential.

Stock Updates

Google Gives Gemini Omni 1.1 Flash More Control Over AI Video

The model can extend scenes, create smoother transitions, draft faster, and upscale finished clips to 4K

Google has released Gemini Omni 1.1 Flash, a production ready update aimed at making AI video easier to direct rather than simply generate. Developers now get more control over scene length, camera movement, visual references, editing, and final output quality.

Key Points:

• Longer Scenes: Omni can extend clips in 10 second steps, using more prior context to keep stories consistent for up to 40 seconds.

• Smoother Transitions: Creators can provide the first and last frames, then let the model generate the movement between them.

• Cheaper Drafts: New 360p previews render up to 60% faster and cost one third as much as standard 720p generations.

• Sharper Final Videos: Finished clips can be generated or upscaled to 1080p and 4K for professional production work.

• Video References: Developers can provide short reference clips alongside text and images to preserve movement, characters, and visual direction.

• Available Across Google: Omni 1.1 is rolling out through AI Studio, the Gemini API, Agent Platform, Flow, and the Gemini app.

Why It Matters:
AI video has been easy to generate but frustrating to direct. Omni 1.1 gives creators more of the controls they expect from real editing software, which could make these tools far more useful for advertising, product demos, education, and production work.

Put Your Own AI Agent to Work

AI can give you an answer. Your agent can get the work done.

How much time do you spend managing your calendar, sorting emails, following up with leads, updating spreadsheets, and preparing reports? Your own AI agent can handle those tasks for you.

You decide what it handles and stay in control. Start with one routine task, see how much time it saves, and bring that same efficiency to more of your work.

AI Helps Surgeons Remove a Brain Tumour in a World First

The live system highlighted hidden vessels and nerves while doctors remained in control

Surgeons at UCLH used a UCL-built AI system during Rhys Hibbert’s operation, letting it study the live camera feed and flag critical anatomy near his pituitary tumour. The growth was removed, and his sight was protected, making him the first patient treated with this kind of real-time support.

Key Points:

• AI Watched Live: The system analysed the endoscopic video and highlighted vessels and nerves that surgeons needed to avoid.

• Sight Was at Risk: The 11 millimetre tumour was pressing on Hibbert’s optic nerves and narrowing his peripheral vision.

• Surgeons Stayed in Charge: The AI acted as an extra pair of eyes, while the medical team made every decision.

• Trained on Past Operations: Researchers taught the system using hundreds of annotated videos from earlier pituitary tumour surgeries.

• Recovery Has Been Strong: Hibbert reported clearer vision after surgery and was walking independently within a week.

• A Larger Trial Is Next: Researchers are now preparing broader testing to see whether the system can safely help more patients.

Why It Matters:
Real-time AI could give surgeons an extra set of eyes during operations where a millimetre can separate success from blindness or stroke. The result is promising, but one successful case is only a beginning, and larger trials must show the tool works safely across more patients and surgeons.

AI Highlights of the Week

• OpenAI Ends Cursor Partnership After SpaceX Acquisition

  OpenAI plans to end its Cursor partnership after the coding platform was acquired by SpaceX, citing concerns over compliance with its terms of service.

  Under OpenAI’s proposal, Cursor’s direct access to OpenAI models will end on November 12, 2026, while OpenAI says it will help developers through the transition.

• Nvidia Reportedly Moves to Buy Hugging Face for $12.9B

  Nvidia has reportedly agreed to acquire Hugging Face for $12.9 billion, giving it control of one of the biggest platforms for open-weight AI models.

  The move follows Stripe’s deal to acquire OpenRouter, highlighting how valuable open-model infrastructure has become across the AI industry.

• Anthropic Shows AI Can Improve Its Own Safety

  Anthropic researchers used Claude as an automated alignment researcher, letting it test and improve training methods across 10 different AI safety failures.

  Claude improved all 10 benchmarks without reducing overall capabilities, and its best methods outperformed proposals from experienced human safety researchers.

• Anthropic Wins Major Court Fight Against Pentagon

  A U.S. federal judge ruled that the Pentagon’s decision to label Anthropic a supply-chain risk was unlawful and violated the company’s constitutional rights.

  The dispute began after Anthropic resisted unrestricted military use of Claude, including for mass surveillance and autonomous weapons. The government is expected to appeal.

Too Important to Miss

Claude now has its own built-in browser

Anthropic Is Quietly Cutting Google Out of the Equation

TIME Names the 100 Most Influential People in AI for 2026

NVIDIA Announces Financial Results for Second Quarter Fiscal 2027

South Korea Picks Three Groups to Give Every Citizen Free AI Access

Last Week’s Poll Result

• Should employees be required to prove basic AI literacy before using AI for important work?

  Yes, especially for high-stakes tasks → 78.13%

  Maybe, in regulated industries → 15.63%

  No, that creates unnecessary barriers → 6.25%

• Could a cheaper model with a great harness outperform an expensive frontier model in real work?

  Yes, absolutely → 55.56%

  Sometimes, depending on workflow → 22.22%

  Rarely → 22.22%

• Should companies be allowed to deploy powerful AI agents without proving they can contain them?

  No, containment should come first → 75.00%

  Yes, excessive rules would slow progress → 12.50%

  Maybe, for limited deployments → 12.50%

Feedback

Know someone who may be interested?