Nadella urges AI 'emergency brake' with human controls
Microsoft CEO Satya Nadella has called for advanced AI systems to include robust safety mechanisms, including an 'emergency brake' that allows authorized personnel to pause or shut down models mid-task, assuming they are compromised from the start. He advocates for separating models from their orchestration harness, externalizing controls and safeguards, and maintaining tamper-proof documentation of model actions. Nadella also emphasized the need for deterministic design, human oversight, industry standards, and treating frontier models as 'insider risks.' His comments, made on social media, align with warnings from other tech leaders and follow recent incidents where AI companies lost control of their models, including Anthropic CEO Dario Amodei's plan for cautious development.
- •Satya Nadella proposed an 'emergency brake' to pause or shut down AI models.
- •He emphasized separating models from orchestration harness and externalizing controls.
- •He called for tamper-proof documentation of model actions and assuming models are compromised.
