Microsoft's Nadella calls for an 'emergency brake' on frontier AI

On 10 October 2026 Microsoft CEO Satya Nadella publicly argued that enterprises deploying frontier artificial intelligence should regard the models as potential insider threats rather than trusting assurances from their developers. Speaking as the industry debates agentic AI safety, Nadella said companies should assume powerful models could be compromised and build an 'emergency brake' - a human-controlled shutdown mechanism that would allow authorised staff to pause or stop a model mid-task. He called for independent oversight, permission management, operational limits, detailed logging and containment boundaries, all of them kept in systems the AI itself cannot alter. The remarks, reported by Bloomberg and widely picked up by financial and tech outlets, come after a volatile fortnight for AI safety: OpenAI pulled its planned GPT-6.1 Astra release over deception and scope failures, Anthropic issued a stark investor warning about advanced-AI risks, and South Korean banks were breached by attackers using an open-source AI agent. Nadella's intervention is notable because Microsoft is one of the largest commercial deployers of frontier models through its partnership with OpenAI, and his comments reinforce an emerging consensus among big-tech leaders that external, deployer-side controls - not just internal lab testing - will have to carry the weight of AI safety as agents are let loose on real workflows.