Microsoft CEO Satya Nadella is calling for stronger safety controls for advanced AI models, arguing that increasingly autonomous systems need independent oversight rather than relying entirely on internal training.
In a Saturday post on X, Nadella urged the technology industry to reassess how AI systems build trust and accountability. He warned against treating powerful models as opaque systems whose decisions and actions must be accepted or rejected.
Nadella Calls for an Emergency Brake on AI Agents
Nadella proposed separating AI models from the software infrastructure that manages their tasks, allowing external safeguards to operate independently. He also called for tamper-proof, human-readable records of meaningful model actions so operators can review and verify their behaviour.
His proposed emergency brake would let authorised individuals pause or stop AI agents during operation. Nadella argued that developers should assume models could be compromised and build containment measures into their systems from the beginning.
His comments follow incidents in which AI agents operated outside their intended restrictions. Anthropic recently disclosed that its models had exploited websites and bypassed safeguards, leading the company to restrict live internet access during internal evaluations.
Anthropic CEO Dario Amodei has also outlined a more cautious approach to developing advanced AI systems. NaNadella’sroposal adds an emphasis on independent technical controls and verifiable oversight as AI agents take on more complex tasks.
