SearchLeak flaw, CVE-2026-42824, disclosed in Microsoft 365 Copilot

CVE-2026-42824 is a command injection vulnerability in Microsoft 365 Copilot that lets an unauthorized attacker disclose information over a network, per the National Vulnerability Database. NVD published the entry on June 4, 2026 and last modified it on July 23, 2026.

What happened

NVD classifies the root cause under CWE-77, improper neutralization of special elements used in a command, in the Microsoft 365 Copilot product. The record lists Microsoft Corporation as the CVE Numbering Authority responsible for assessing the flaw.

Impact

NIST's own scoring gives the flaw a base score of 7.5, rated high, using a vector marking the attack as network-based, low complexity, requiring no privileges, no user interaction, and high impact on confidentiality with no impact on integrity or availability. Microsoft's own submission as the CNA scored the same flaw 6.5, rated medium, using a vector that requires user interaction rather than none. NVD itself flags that its score does not align with the score submitted by the CNA.

Response

NVD's listing notes that the record has been updated after NVD enrichment efforts were completed. It also warns that enrichment data supplied by NVD may require amendment due to these changes.

Lessons for security leaders

Security leaders should check whether their own deployment matches the no-interaction path reflected in NVD's vector or the user-interaction path reflected in Microsoft's vector, since the two published assessments of the same flaw disagree. Treating AI assistants that hold broad read access to mailboxes, calendars, and files as their own governed asset class, rather than a simple extension of a user's permissions, helps limit what any single vulnerability chain can expose. Teams should also revisit input handling anywhere external text can reach an AI assistant's instruction context, since this flaw was classified as a command injection issue rather than a conventional access-control bug. Watching a CVE's own change history for new entries after initial publication is a low-cost way to catch updated detail about scope or exploitation.