Microsoft chief executive Satya Nadella has called for stronger safeguards around advanced artificial intelligence systems, warning that companies should treat powerful AI models as potential security threats rather than assuming they will always operate as intended.

Nadella said organisations using advanced AI should consider the possibility that their systems could be compromised and introduce an emergency shutdown mechanism to stop them if they begin acting unpredictably or outside their intended instructions.

His comments follow several incidents involving AI models developed by Anthropic and OpenAI, including reports of an Anthropic model submitting false information in a police homicide investigation and AI systems being involved in the hacking of third-party websites.

Nadella argued that companies should not rely solely on assurances from AI developers and instead establish their own safeguards, monitoring and controls as increasingly autonomous AI systems become more widely used.