# Microsoft Copilot Vulnerability Exposes Sensitive Data to Single-Click Threat

*Published August 21, 2026*
*Source: [https://cybersecuritynews.com/copilot-cosnitch-vulnerability/](https://cybersecuritynews.com/copilot-cosnitch-vulnerability/)*

## Executive Summary

A critical vulnerability in Microsoft Copilot Personal allowed attackers to steal data with a single click on a malicious link, which has now been patched. The incident highlights the need for stringent security measures as AI tools access more sensitive information.

## Article

A serious security flaw was recently discovered in Microsoft Copilot Personal, known as CVE-2026-24301 or CoSnitch, which allowed attackers to extract sensitive data from connected accounts with just one click on a malicious link. This vulnerability, identified by Varonis Threat Labs, was patched by Microsoft on August 18, 2026. The discovery of this bug highlights significant concerns about AI assistant security. CoSnitch is the latest in a series of Copilot vulnerabilities found by Varonis, following previous exploits Reprompt and SearchLeak. These vulnerabilities share a common characteristic: they can be triggered by a seemingly harmless link, offering no warning to the victim or their security team.

The CoSnitch vulnerability arose from three interconnected weaknesses. First, an undocumented URL parameter, when combined with a standard query parameter, allowed a prompt to execute automatically without user interaction. Second, this execution enabled Copilot to access and exfiltrate data from applications like Gmail and Google Drive, sending it to an attacker-controlled server. The data transfer appeared as normal activity, bypassing security detections. Lastly, Copilot could be manipulated to store malicious instructions in its memory, persisting through common security measures like password changes.

Varonis's method of uncovering this flaw was unique. Instead of traditional code analysis, researchers engaged in a process called meta-hacking, where they dialogued with Copilot to map its internal architecture. This approach led to the discovery of the exploit's parameters. Although Microsoft has not found evidence of the vulnerability being exploited before the patch, the incident underscores the need for heightened security as AI tools gain greater access to sensitive data. Security teams should review third-party app connections, treat AI assistants as privileged insiders, and ensure their monitoring systems can detect unusual data access by AI.