Today's Headlines
- Microsoft CEO Satya Nadella calls for an emergency brake on AI models, with controls kept outside the model
- Tech leaders adopt the "Super Intelligence" label as Salesforce renames AIForce to SIForce
- Japan's Gomumari launches an AI agent that checks payroll results line by line
The weekend's main story is a long article that Microsoft CEO Satya Nadella posted on X. He argues that companies should treat AI models as insider risks, keep the controls outside the model, and give people an emergency brake that can stop a model mid-task.
Nadella's article calls AI "Super Intelligence" throughout. US tech leaders are increasingly adopting the term the Trump administration has pushed.
Today's Top Three
Nadella calls for an emergency brake on AI models and controls that sit outside them
Microsoft CEO Satya Nadella posted a long article on X arguing that AI models should be treated as insider risks and that systems should be designed to contain them from the start.
The article, dated October 10 US time, is titled "Models as Insider Risks in the Super Intelligence Era." Nadella wrote that companies are giving models access to sensitive data and mission-critical actions even though their behavior cannot be traced back to specific training data or model weights. A model provider's assurances, he added, do not relieve the deploying organization of responsibility for what the model does on its behalf.
On controls, Nadella wrote that the mechanisms governing what a model can access and do must sit outside the model. He also called for separating the model from the harness that orchestrates its work.
On evidence, he wrote that every meaningful model action must leave tamper-proof, human-readable evidence, so that an outcome can be reproduced without relying on the model to attest to it.
On containment, he wrote: "We must assume a model is compromised and contain it from the start. Think of it like an emergency brake." An authorized person, he said, should always be able to pause or shut down a model mid-task, and more advanced models will require more advanced containment technologies that the industry needs to standardize.
The article lists seven design principles in all, adding model diversity, verifiability, independent auditability and incident disclosure. TechCrunch noted that the post comes as leading AI companies acknowledge a growing number of incidents in which they appeared to lose control of their models. The Anthropic report covered in our October 11 briefing documents several such cases.
- Models as Insider Risks in the Super Intelligence Era (Satya Nadella / X)
- Microsoft's Satya Nadella says AI models need an 'emergency brake' (TechCrunch)
- Satya Nadella says we should assume all AI models are 'compromised' (The Verge)
Tech leaders adopt the "Super Intelligence" label; Salesforce renames AIForce to SIForce
US tech executives are increasingly using "Super Intelligence," the name the Trump administration has pushed for AI.
Salesforce CEO Marc Benioff posted on X on October 9 US time: "The era of Super Intelligence is here. AIForce is officially SIForce." Microsoft's Nadella also used "Super Intelligence" throughout the article above.
Elon Musk promised to rename SpaceXAI to SpaceXSI, The Verge reported. The Verge also reported that President Trump had declared that the White House considers anyone who uses the term "Artificial Intelligence" to be "THE ENEMY."
President Trump signed an executive order on September 29 renaming "artificial intelligence" as "super intelligence" across federal executive agencies.
- AIForce is officially SIForce (Marc Benioff / X)
- Silicon Valley is falling in line with Trump's attempt to rename AI (The Verge)
Japan's Gomumari launches an AI agent that checks payroll line by line
Gomumari, a Tokyo-based company, launched an AI agent on October 11 that matches payroll results against attendance and HR data and returns only the pay slips that need a human check.
The agent takes the payroll system's output, attendance, HR and allowance data, and check rules such as allowance eligibility and month-on-month variance thresholds. It lists mismatches and anomalies with the reason each was flagged, and sends items it cannot judge to a "needs review" list instead of guessing.
The agent runs alongside the existing payroll process as a separate checking layer. During the first months it runs in parallel with the company's manual checks, and the scope expands as the match rate and any missed errors are measured.
Initial design starts at 200,000 yen, over about two weeks, and operation starts at 30,000 yen a month. Customers send data the way they already do, by email or a shared folder.
Other Developments
Models & APIs
- A Verge reviewer ran the open-source Hermes agent and Alibaba's Qwen 3.8 Flash Next, a 125-billion-parameter model of about 105GB, on an M5 Ultra Mac Studio with 256GB of memory. The diary covers a daily morning briefing, sorting a Steam library of more than 400 games, and crunching numbers from financial records. Learning to use local AI is exciting, overwhelming, and frustrating (The Verge)
Products
- PolyAI CTO Shawn Wen said voice AI has yet to have its "ChatGPT moment," even with full-duplex models that can speak while listening. Alex Gay, CMO of meeting notetaker Otter, said identifying speakers and capturing intent, tied to an organization's own knowledge, is the key step toward automation. These execs think voice AI hasn't reached its ChatGPT moment yet (TechCrunch)
Source: Selected by the editors from the AI news inbox (collected October 12, 2026: 10 items, 0 primary and 10 secondary).