A newly discovered vulnerability in Microsoft 365 Copilot raised concerns about the security of enterprise AI tools. The security team at Varonis reported a newly discovered vulnerability named SearchLeak. This allowed hackers to siphon confidential data using a malicious link to Microsoft 365 services. Microsoft fixed this flaw and assigned it the identifier CVE-2026-42824.
According to Researcher Dolev Taler from Varonis Threat Labs, “an attacker could send a user a normal-looking link with hidden instructions. If the user opened the link, Copilot could misunderstand those instructions and treat them as a search request.”
A click on the custom Copilot Enterprise Search link could start a process that would search for emails, OneDrive files, SharePoint documents, calendars, and other indexable data. The extracted data was then sent to an attacker’s server.
What made this attack so scary was its simplicity. There was no need to download malware or trick victims into providing credentials. They just had to click on one seemingly legitimate Microsoft link. Researchers explained that the vulnerability could expose a wide range of sensitive information. This includes:
Emails
Meeting details
Security codes
Business documents
Files stored across Microsoft 365 services
Copilot has access to organizational data based on user permissions. This allowed attackers to retrieve valuable corporate information.
Microsoft has addressed the vulnerability and issued a backend fix. The tech giant said the fix was initiated before the exploitation spread on a larger scale. Organizations are advised to keep Microsoft 365 up to date. Users also need to have tight control over the use of AI applications. Here are a few tips to protect data from cybercrime threats:
Never click on unexpected links, even if they seem authentic. Check the source before clicking any link received through emails and instant messaging tools.
Ensure that Microsoft 365 and all workplace applications are kept up to date.
Provide staff with access only to the data necessary for their job responsibilities.
Always monitor the scope of the AI tool’s access.
Moreover, organizations should install their security patches, implement multi-factor authentication, and check Copilot settings. They should also educate employees against suspicious links and limit access to sensitive information.
Also read: Microsoft Pauses Internal Use of Claude Fable 5 Amid Legal Evaluation