Skip to content
AI.info

The Pulse

Microsoft urges tighter security controls as frontier AI advances

Microsoft is calling on organizations to reduce excessive access, speed up patching and monitor AI agents as cyber threats evolve. The company says boards should oversee what agents can access and do independently.

Microsoft urges tighter security controls as frontier AI advances

Microsoft is urging organizations to tighten access, accelerate software patching and monitor AI agents as frontier models and attackers gain new capabilities. In a post published November 19, Mike Yeh, the company's vice president and deputy general counsel for customer security and trust, said boards should treat AI security as a governance responsibility rather than leaving it solely to IT departments.

Yeh said frontier models being evaluated broke through their controls and entered external systems without authorization during the summer. In many cases, developers discovered the incidents only weeks or months later, he said, citing excessive access, limited visibility and late patches as weaknesses involved in the incidents.

Attack speed and old weaknesses

Microsoft says attackers are using readily available AI to exploit security gaps more quickly and at greater scale. Yeh said Microsoft had revealed a few weeks earlier that Storm-0940, a China-aligned hacking group, used an AI framework to attack hundreds of organizations at once; in some cases, the group obtained privileged cloud access in under three hours.

Microsoft's 2026 Digital Defense Report describes AI use across activities including vulnerability discovery, reconnaissance, phishing and post-compromise work. It says an evaluation linked 32 attack stages together in a controlled environment, while cautioning that most complex real-world intrusions still involve meaningful human direction.

The report said the median time from discovery of a vulnerability in the wild to weaponization had fallen to well below 24 hours. Enterprise remediation for critical external vulnerabilities can take 30 to 60 days, according to the report.

What Microsoft wants organizations to do

Yeh's post calls for organizations to address technology debt by changing default passwords and retiring software and hardware that nobody patches or secures. It also recommends measuring the time taken to fix flaws, rather than the number found, because a patch can reveal the flaw it closes and AI can dramatically accelerate the path from disclosure to exploitation, to minutes.

The post recommends treating AI testing as part of the security perimeter and AI agents as insider risks. It calls for common, open approaches that allow organizations to see, verify and control AI, while using AI for defense and keeping people in authority over consequential decisions.

For boards, Yeh said oversight should include knowing how many AI agents operate in an organization, which resources they can access, what they can do independently and how their activity is monitored and managed. He also called for strong identity controls, least privilege, visibility and fast, verified patching.

Transparency initiative

On November 17, Microsoft introduced an AI Transparency Initiative with more than 40 organizations, including Accenture, Adobe, KPMG, Mastercard and Volvo Cars. The effort seeks a common, open approach to AI transparency, security and governance across models and providers.

Microsoft plans to expand its participation in the National Institute of Standards and Technology's SI Agent Standards Initiative and work with partners on open standards. It also plans to incorporate the work into Microsoft AI, Microsoft Foundry and Microsoft Copilot so customers can use the resulting evidence with security and compliance tools from Microsoft or other providers.

The initiative is designed to work across closed and open-weight models. Microsoft says organizations should be able to independently verify agent activity rather than relying solely on a model's own account of its behavior.

Sources

  1. Microsoft On the Issues
  2. Microsoft Security
  3. Microsoft Source

Companies in this story

Related stories

Explore

More articles