AI agents must be given limited access to network: Microsoft V-P Mike Yeh

Microsoft's Mike Yeh warns that enterprises must restrict AI agents' access to sensitive systems, as growing adoption and faster AI-driven attacks heighten cybersecurity risks

Companies deploying artificial intelligence (AI) agents must restrict their access to enterprise systems and sensitive data to prevent them from overstepping their assigned tasks, a senior Microsoft executive has warned.  

Mike Yeh, vice-president and deputy general counsel of customer security and trust at Microsoft, believes that enterprises must give least amount of privilege and access of their networks to contain them and stop their overreach. 

“When you look at how consumers are using agents today… they are giving them access to their email, bank accounts. You can see why that will create problems in the near term. As we get smarter about that, we have to change the mentality. The default is the least amount of privileges for the agent to do its task,” Yeh told Business Standard in an interaction in New Delhi. 

There should also be more transparency from frontier AI-model developing companies about the capabilities of the AI agent they are releasing to users, so that “there is a better understanding of what the consequences could be,” Yeh said. 

Speaking on the 2026 Microsoft Digital Defense Report, Yeh said that while most cyberattacks still exploit old vulnerabilities, going forward, everything will have to be seen through the lens of how AI has impacted it. 

“We have a very small window right now to get the basics right. For example, vulnerabilities from the year 2020 that were exploited this year should have been patched. This is where the industry needs to do its homework before people start using AI in a way that really compresses the attack cycle timeline for us,” he said. 

To give enterprises and customers more options and control over their digital networks, Microsoft has also introduced newer control options for Azure users, Yeh said. 

On Tuesday, Microsoft’s chief executive officer Satya Nadella announced in Paris a host of new initiatives from the company on expansion of sovereign Cloud capabilities, including giving Azure users a unified way to assess their sovereignty posture, identify the gaps in their existing Azure capabilities, apply the necessary guardrails as per the rules of the country of operation, and retain evidence of qualifying Microsoft operator access. 

Speaking on the newer initiatives announced by Nadella on Tuesday, Alistair Speirs, the general manager at Microsoft Azure, said different countries now have varying definitions of sovereignty, so companies do not have a mandatory set list of standards they must meet. 

“So, we really think of sovereignty as a risk management exercise. And as a risk management exercise, it sits alongside other risks, including business continuity, natural disasters, and cybersecurity,” Speirs said during a select media roundtable.