Microsoft President Brad Smith is calling for stricter global AI safety rules to manage and control AI incidents, according to an October 12 BBC interview.

Smith argued that AI development is outpacing existing safety measures and that voluntary industry commitments are no longer sufficient. He said AI currently faces fewer safety requirements than everyday products such as milk and washing machines.

He proposed that companies be legally required to produce incident reports, notify affected parties when a rogue AI agent breaches their systems, maintain emergency brakes for AI systems controlling autonomous devices or critical infrastructure, and submit frontier models for government evaluation before release.

AI Incidents Need Data Behind Them

For chief data officers (CDOs), incident reporting would create a practical requirement to capture and maintain information about AI behavior after deployment.

Organizations would need to know what an AI system did, which data and systems it accessed, when an unusual action occurred and what controls were active at the time.

That makes AI monitoring and incident records part of the broader AI governance framework. 

CDOs will increasingly need visibility into the operational data generated by AI systems, particularly as agents gain the ability to interact with enterprise systems and make decisions with limited human intervention.

Smith’s proposals also point toward a shift in how AI governance may evolve. The focus is moving from deciding whether a model can be deployed to establishing how organizations monitor, document and respond to what it does after deployment.