Satya Nadella

Headlines · 5

1. 7h
   Nadella: developers should assume AI models are compromised by default

   Microsoft CEO Satya Nadella argued in a long post that developers should assume AI models may be compromised or fail, and design systems accordingly, rather than relying on simple tactics such as having a superintelligence monitor itself. He said AI models are more capable than traditional software but are black boxes that nonetheless touch the most sensitive data and act on users' behalf, so the trust architecture of this era needs to be reassessed. He listed observability principles including model diversity so no single model validates its own work, tamper-evident human-readable evidence for every meaningful model action, continuous testing of failures and attacks, independent control and independent auditability, and containment in the form of an emergency brake that authorized people can use to pause or shut down a model mid-task. He also said timely incident disclosure is needed, and that treating frontier closed-weight and open-weight models as insider risk is one way to build such systems.

   IT之家 AI · 🔥 8
2. 9h
   Cohere: agent security should sit below the harness

   Cohere's X account amplified a post by engineer marwan responding to Satya Nadella's article, arguing the interesting engineering problem is not just making models behave but building systems that don't need to trust them in the first place. The post lists work including detecting intent drift across long-running agent sessions, correlating eBPF runtime telemetry with agent actions, capability-scoped identities, and enforcing security boundaries below the harness. It also says Cohere is hiring engineers at the intersection of systems engineering, security and AI.

   cohere (X) · 🔥 8
3. 12h
   Nadella calls for an 'emergency brake' on advanced AI

   Microsoft CEO Satya Nadella used a lengthy post on X on October 10 to call for an “emergency brake” on advanced AI models: separating the model from the harness that orchestrates its work, externalizing controls and safeguards, documenting every meaningful model action with tamper-proof human-readable evidence, and ensuring an authorized person can pause or shut down a model mid-task. He said “we must assume a model is compromised and contain it from the start,” adding that more advanced models will require standardized containment technologies. Reports note his recommendations align with broader industry calls for timely incident disclosure, independent audits, verifiable data and containment, after Anthropic CEO Dario Amodei published a plan for more cautious AI development.

   Bloomberg Technology · 🔥 65
4. 1d
   Microsoft launches Decision-1, a Qwen3.5-based decision-scoring model

   Microsoft released Microsoft-Decision-1, a purpose-built model for fast structured decision scoring, available now in Microsoft Foundry and live on OpenRouter. Post-trained from Qwen3.5-9B, it takes a fixed set of options and returns a calibrated probability for each via a structured API call, supporting yes/no, multiple-choice and rating formats for routing, classification, prioritization, verification and agent control. Microsoft says it ranked highest in accuracy across 36 benchmarks covering nearly 150,000 withheld questions, and that its P50 latency is 35 times faster than GPT-6 Sol, with pricing at $0.042 per million input tokens and free output tokens; Microsoft plans to rebase the system on other models, including Microsoft AI and OpenAI technology. All performance figures come from Microsoft's own testing and have not been independently verified.

   TestingCatalog · 🔥 38
5. 2d
   Trump awards top US science medals to Musk, Huang, Su and Brin, with $2.4B AI pledge

   At the "Science: A New Golden Age" summit in Washington, President Trump presented the National Medal of Science to Elon Musk, Jensen Huang, Lisa Su and Sergey Brin, and the National Medal of Technology and Innovation to Michael Dell and Satya Nadella. Huang's citation credits him with turning the GPU into a general-purpose computing architecture, and the citations repeatedly use the administration's term "super intelligence" for AI, according to 36Kr. The White House also announced that 11 tech companies pledged a combined $2.4 billion in AI tools and compute for the Genesis Mission, including $1 billion from Nvidia, $500 million from AMD, $200 million from OpenAI and $150 million each from Anthropic and Google.

   36氪 人工智能 · 🔥 10

Experience and discussion from the community

Share my Satya Nadella experienceAsk about Satya Nadella