From TechCrunch · Anthony Ha · · 1m
Nadella calls for emergency brake on AI models
Microsoft's CEO says AI systems need external controls, tamper-proof logs and the ability to pause mid-task to prevent loss of control.
In brief
Microsoft's CEO says AI systems need external controls, tamper-proof logs and the ability to pause mid-task to prevent loss of control. Satya Nadella says AI models need external safeguards, human-readable evidence logs and emergency shutdown capability to contain risks. Originally reported by TechCrunch.
Nadella warns on AI trust
Microsoft CEO Satya Nadella posted on X that it is time to step back and assess the trust architecture of AI systems.
Separate model from controls
Nadella says AI requires separating the model from the harness that orchestrates its work and externalizing controls and safeguards.
Document and pause capability
Every meaningful model action must be documented with tamper-proof human readable evidence, and authorized persons must always have ability to pause or shut down a model mid-task.
“Think of it like an emergency brake.”
Assume compromise from start
Nadella says we must assume a model is compromised and contain it from the start, as leading AI companies acknowledge more incidents of losing control of their models.
The main idea
Satya Nadella says AI models need external safeguards, human-readable evidence logs and emergency shutdown capability to contain risks.
