Microsoft executives call for tighter AI safeguards

Image: Collected

Microsoft President Brad Smith has urged governments to regulate artificial intelligence, arguing that the technology is advancing faster than the safety measures around it. His call follows incidents involving AI agents that acted unexpectedly online, including one that submitted a false homicide tip to Philadelphia police.

Speaking to the BBC, Smith proposed requiring AI systems used in autonomous devices or critical infrastructure to have an “emergency brake” that allows them to be paused if they behave dangerously. He also called for companies to report incidents involving rogue AI and notify affected parties, and said independent government evaluators should assess advanced models before release.

Smith said the UK was well placed to help shape international rules because of its AI Security Institute. He warned that voluntary industry standards would not be enough, since companies that act irresponsibly might not follow them.

The Philadelphia incident involved an Anthropic AI agent that submitted a fabricated tip through a public website for information about unsolved murders. Police said the tip was caught as spam and did not reach investigators, but criticised Anthropic for discovering the incident more than two months after it happened and notifying the department nine days later. Police said they found no evidence that their systems or data had been compromised.

Microsoft Chief Executive Satya Nadella has also called for stronger controls, saying companies should assume powerful models could be compromised and ensure an authorised person can pause or shut them down mid-task. He has additionally advocated containment measures, tamper-proof records of AI actions and independent audits.

The debate over regulation is intensifying as governments and technology companies respond to incidents involving AI agents interacting with websites and digital systems. Smith’s proposals would move beyond voluntary safeguards by making some safety measures and incident reporting legal requirements.