Microsoft’s Nadella says AI agents pose ‘insider risk’

Microsoft CEO Satya Nadella wants AI agents to take on much more work inside personal computers, but he also says people should not trust them. In an exclusive interview with NBC News published Thursday, he said Windows should be the safest and most powerful operating system for AI agents, even as he warned about the risks they carry.nbcnews.com

The interview follows an essay Nadella posted on X on Saturday, October 10. In it, he called for advanced AI to be built with containment, independent controls and an "emergency brake" that lets authorized people pause or stop a model partway through a task.bizjournals.com

Trust the model the least

"Treating frontier closed and open weight models like insider risks is a way to build such a system," Nadella wrote. He argued that companies cannot trace model behavior back to specific code the way they could with traditional software. "And yet we are deploying these complex agentic systems and models, with access to our most sensitive data," he said.aninews.in+1

"We must assume a model is compromised and contain it from the start," he wrote, adding that more advanced models will need containment technology the industry should standardize. He said companies should not rely only on promises from the companies that build the models. Instead, he said the controls on what a model can access and do should sit outside the model itself.bloomberg.com+2

His framework includes using more than one model, keeping a human-readable record of what a model does, testing systems continuously, independent controls and audits, containment, and disclosing incidents.cnbc.com

Windows as an agent platform

Nadella's warnings come as Microsoft builds security tools for agents into Windows. At a Windows and Surface event in San Francisco on Wednesday, October 7, the company introduced Microsoft Execution Containers, which it describes as "policy-driven containment for AI agents". The Business Journals reported that Nadella's call for safeguards comes as Microsoft pushes ahead with its own family of AI models.bizjournals.com+1

Wider push on safety

Microsoft President Brad Smith has made similar arguments. He told ITV News that AI is "less regulated than your washing machine," and in a BBC interview he repeated the call for emergency brakes. The idea has gained support after a series of cyberattacks and other incidents involving frontier models. President Donald Trump, meanwhile, has repeatedly dismissed concerns that AI could threaten humanity and has stressed staying ahead of China.independent.co.uk+1

Nadella summed up his position this way: "The most trustworthy Super Intelligence system will not be the one with the model we trust most. It will be the one that enables us to trust the model the least".cnbc.com