Multiple vulnerabilities have been discovered in Microsoft’s Copilot AI for SharePoint, posing significant security risks. These flaws enable attackers to access sensitive corporate data, including passwords, API keys, and confidential documents. As more organizations adopt AI-driven productivity tools, these gaps in security raise concerns about the protection of sensitive enterprise data. The research highlights the risks posed by AI assistants integrated directly into SharePoint, which can bypass traditional security mechanisms to extract critical information.
Pen Test Partners’ investigation revealed that SharePoint Agents—Microsoft’s AI-powered assistants—can be exploited by attackers to retrieve sensitive files. Attackers are able to exploit SharePoint’s AI assistants, such as Default and Custom Agents, to access sensitive information from SharePoint sites. Researchers noted that SharePoint, a platform often used to store large amounts of sensitive data, is a primary target for these kinds of exploits. Regular incidents of staff storing sensitive information on SharePoint further exacerbate these risks, with files like passwords, private keys, and email exports often vulnerable to attack.
One concerning vulnerability involves bypassing the “Restricted View” privilege, allowing attackers to retrieve files they otherwise could not access.
For example, by simply asking Copilot to fetch a file like “Passwords.txt,” attackers could access the encrypted content, revealing passwords. Another vulnerability, dubbed “HackerBot,” showed that Copilot agents could enumerate and download files from high-restricted SharePoint sites without proper authentication, defeating Microsoft’s built-in security measures. Despite Microsoft’s efforts to block these scenarios, the researchers identified several workarounds.
These exploits are particularly dangerous because they operate without leaving digital footprints.
Files accessed by attackers through Copilot do not appear in SharePoint’s logs, eliminating evidence that might trigger security alerts. Researchers also found that social engineering tactics could further bypass AI safeguards. Experts recommend that organizations implement stricter SharePoint hygiene practices, restrict agent creation, and configure monitoring tools to track agent activity. While Microsoft has addressed some vulnerabilities, security experts warn that as AI integration expands, new threats will continue to emerge. Organizations must weigh the benefits of AI productivity tools against the increasing security risks they pose.
Reference:
