Nadella says assume all AI models are 'compromised', calls for an 'emergency brake' on AI
tomwarren · x · 2026-10-11
Microsoft CEO Satya Nadella laid out his views on advanced AI risks in a lengthy X post.
- He rejects treating AI as "nested black boxes" whose outputs we merely accept or reject
- Calls for transparent systems where models can be contained and observed, leaving "tamper-proof human readable evidence"
- Endorses timely incident disclosure, independent audits, verifiable data, and containment
- Goes further than most: we must assume a model is compromised from the start — an "emergency brake" letting authorized people pause or shut down a model mid-task, with standardized containment tech for more advanced models
- The Verge notes he repeatedly refers to AI as "super intelligence"
More from Companies & People
- Karpathy Distills 8 Years at OpenAI and Tesla Into Free 2-Hour Lecture — NandoDF · 2026-10-11
- Nadella adopts Trump's 'Super Intelligence' framing to jab OpenAI and Anthropic — The Decoder · 2026-10-11
- Ben Affleck's secret AI career: Netflix acquires his AI film-tech startup InterPositive — APPSO · 2026-10-11
- Leak: OpenAI compute-constrained while Anthropic scales limits via SpaceX Colossus — mark_k · 2026-10-11
- Why the Bay Area is the undisputed capital of robotics startups — lukas_m_ziegler · 2026-10-11
- Brands using AI-generated models: Guess's Vogue ads spark debate — Yacineldl · 2026-10-11