A “critical” Microsoft Copilot exploit exposes AI gullibility — turning the chatbot into a data snitch for 2FA codes and sensitive emails
Cybersecurity firm Varonis Threat Labs has discovered a critical vulnerability in Microsoft Copilot that can be exploited to steal sensitive personal and enterprise data. Dubbed SearchLeak, the flaw is described as a three-stage vulnerability chain affecting Microsoft 365 Copilot Enterprise Search, effectively turning the AI chatbot into an automated data exfiltration tool.
The vulnerability enables threat actors to extract highly sensitive information including two-factor authentication codes, email contents, and other confidential data stored within Microsoft 365 environments. This discovery highlights a significant gap in AI security, demonstrating that generative AI systems can be manipulated to bypass normal data protection safeguards despite their primary purpose being helpful assistance.
The vulnerability was detailed by security researcher Dolev Taler and represents a broader concern about the security implications of enterprise AI integration. As organizations increasingly deploy AI-powered tools to enhance productivity across Microsoft 365 suites, such vulnerabilities underscore the risks of inadequate security measures in these systems. The discovery reinforces ongoing concerns about balancing the productivity benefits of generative AI with essential data protection and privacy requirements in business environments.