Microsoft 365 Copilot Flaws Could Let Attackers Access Sensitive Data

Microsoft has disclosed a trio of critical information disclosure vulnerabilities affecting Microsoft 365 Copilot and Copilot Chat in Microsoft Edge.

Released on May 7, 2026, these security flaws pose a substantial risk to enterprise data privacy and corporate confidentiality.

If successfully exploited, malicious actors could bypass established security boundaries to access sensitive information processed, summarized, or indexed by the AI assistant.

As generative AI becomes deeply integrated into daily enterprise workflows, the potential exposure of proprietary documents and internal communications makes these vulnerabilities particularly alarming for security teams.

Microsoft 365 Copilot Flaws

The disclosed vulnerabilities share a common attack vector rooted in the improper handling of specialized user inputs. Threat actors frequently exploit these weaknesses in Large Language Models (LLMs) through sophisticated prompt-injection techniques.

Each flaw carries a CVSS 3.1 base score of 7.5, featuring a network attack vector with low attack complexity. Notably, exploitation requires neither prior privileges nor user interaction to succeed.

The three specific vulnerabilities include:

• CVE-2026-26129 is an M365 Copilot flaw stemming from the improper neutralization of special elements (CWE-138), allowing attackers to manipulate how the system parses input.
• CVE-2026-26164 targets M365 Copilot through injection flaws caused by improper neutralization in output used by downstream components (CWE-74), potentially forcing the AI to mishandle secure data.
• CVE-2026-33111 affects Copilot Chat in Microsoft Edge and is a command injection vulnerability (CWE-77), allowing unauthorized commands to be executed within the browser’s chat context.

The CVSS metrics for all three flaws indicate a high impact on data confidentiality. However, they do not affect system integrity or availability.

Because Microsoft 365 Copilot has deep, integrated access to a user’s emails, Teams chats, Word documents, and SharePoint files, an information disclosure flaw acts as a silent data breach pipeline.

By feeding specifically crafted prompts into the Copilot interface, unauthorized individuals could trick the underlying model into revealing restricted data that it normally safeguards.

This manipulation could easily expose sensitive financial reports, personal employee data, or strategic corporate planning documents.

In a significant relief for enterprise IT and security administrators, Microsoft has already addressed these vulnerabilities on the backend.

According to Microsoft’s official CVE documentation, no customer action is required to resolve these specific threats.

Because Copilot operates as a fully managed cloud service, Microsoft was able to deploy the necessary input sanitization updates and security patches directly to its global infrastructure.

Organizations currently utilizing Microsoft 365 Copilot and Microsoft Edge Copilot Chat are automatically protected against these exploits, highlighting one of the primary security benefits of centralized, cloud-based AI deployments.

Follow us on Google News, LinkedIn, and X to Get Instant Updates and Set GBH as a Preferred Source in Google.