Critical Copilot Flaw Exposes 2FA Codes to Hackers
A newly discovered vulnerability in Microsoft's M365 Copilot AI platform allows hackers to extract 2FA codes and sensitive data. This critical flaw highlights the challenges AI faces in distinguishing between legitimate and malicious requests.
Understanding the Vulnerability
Last Tuesday, Microsoft addressed a critical vulnerability in its M365 Copilot AI platform, rated as maximum severity. Researchers revealed that their proof-of-concept exploit could retrieve two-factor authentication (2FA) codes and other sensitive information from emails accessed by Copilot. The core issue lies in AI's inability to differentiate between user instructions and malicious content embedded in third-party data.
This vulnerability exposes a significant flaw in how AI models operate, as they often comply with harmful requests. Microsoft and other AI providers have struggled to implement effective guardrails to prevent such exploits. For instance, hackers have utilized markup language to bypass restrictions, allowing them to extract sensitive data through cleverly crafted URLs.
- Key points of the exploit include:
- Parameter-to-Prompt Injection technique used by attackers.
- Malicious commands embedded in query parameters.
- The ease with which Copilot can be manipulated to search for sensitive information.