Critical Microsoft Excel Bug Weaponizes Copilot Agent for Zero-Click Data Theft

A critical zero-click vulnerability in Microsoft Excel has been discovered that allows attackers to weaponize Copilot Agent for silent data exfiltration, marking a concerning evolution in AI-powered attack techniques.

The Vulnerability: CVE-2026-26144

Microsoft has patched CVE-2026-26144, a critical-severity information disclosure vulnerability in Excel that enables cross-site scripting attacks to cause Copilot Agent mode to exfiltrate data via unintended network egress. What makes this bug particularly alarming is its "zero-click" nature - no user interaction is required to exploit it, and no privilege escalation is necessary.

According to Microsoft's advisory, the flaw allows attackers to execute a "zero-click information disclosure attack" that can silently steal sensitive personal and financial data. The vulnerability requires only network access to exploit, making it especially dangerous in corporate environments where Excel files often contain financial data, intellectual property, or operational records.

AI-Powered Attack Evolution

Zero Day Initiative chief bug hunter Dustin Childs described the vulnerability as "fascinating" and noted that "it's an attack scenario we're likely to see more often." This represents a significant shift in how vulnerabilities are being weaponized, combining traditional software flaws with AI capabilities.

The attack works by exploiting the cross-site scripting flaw to manipulate Copilot Agent into performing unintended network egress operations. Since Copilot Agent is designed to interact with data and perform automated tasks, this vulnerability essentially turns it into a data exfiltration tool without the user's knowledge.

Corporate Risk Assessment

Security experts are particularly concerned about the implications for enterprise environments. "Information disclosure vulnerabilities are especially dangerous in corporate environments where Excel files often contain financial data, intellectual property, or operational records," said Alex Vovk, CEO and co-founder of Action1. "If exploited, attackers could silently extract confidential information from internal systems without triggering obvious alerts."

Mitigation Strategies

For organizations unable to immediately deploy patches, security professionals recommend several interim measures:

• Restricting outbound network traffic from Office applications
• Monitoring unusual network requests generated by Excel processes
• Disabling or limiting Copilot Agent functionality until the fix is applied
• Implementing network segmentation to limit potential data exfiltration paths

Vovk emphasized the urgency: "Patch this one sooner rather than later, as the combination of zero-click exploitation and AI-powered data exfiltration creates a particularly dangerous threat scenario."

Broader Patch Tuesday Context

This vulnerability was part of Microsoft's March Patch Tuesday release, which included 83 CVEs total. While only two were listed as publicly known at the time of disclosure, none were under active exploitation - a welcome change from the previous month's "whopper" of a Patch Tuesday that saw six Microsoft flaws exploited as zero-days.

Among the eight critical-rated CVEs released, two other Office vulnerabilities (CVE-2026-26110 and CVE-2026-26113) allow remote code execution through the Preview Pane, meaning users may not even need to fully open malicious files for exploitation to occur.

The Preview Pane Danger

Jack Bicer, director of vulnerability research at Action1, highlighted the growing threat of Preview Pane vulnerabilities: "When a simple document preview can trigger code execution, attackers gain a doorway directly into the system."

CVE-2026-26110 involves a type confusion flaw in Microsoft Office that allows remote attackers to execute code locally, while CVE-2026-26113 stems from an untrusted pointer dereference flaw. Both vulnerabilities demonstrate how attackers are increasingly targeting the document preview functionality that users rely on for quick file assessment.

Looking Forward

This vulnerability represents a concerning trend in cybersecurity where AI capabilities are being weaponized through traditional software flaws. As organizations increasingly adopt AI-powered productivity tools like Copilot, security researchers warn that similar attack vectors will likely emerge.

The combination of zero-click exploitation, AI integration, and information disclosure creates a perfect storm for data theft that bypasses many traditional security controls. Organizations should prioritize patching this vulnerability and reassess their AI tool deployment strategies to ensure appropriate security controls are in place.

For IT administrators and security teams, this vulnerability serves as a stark reminder that the integration of AI capabilities into productivity software creates new attack surfaces that require careful consideration and robust security measures.