Description

Microsoft has disclosed three critical information disclosure vulnerabilities affecting Microsoft 365 Copilot and Copilot Chat in Microsoft Edge. Released on May 7, 2026, the vulnerabilities could have enabled attackers to bypass security protections and gain unauthorized access to sensitive enterprise information processed by the AI-powered assistant. The flaws are associated with improper handling of specially crafted user inputs, a technique commonly linked to prompt-injection attacks targeting Large Language Models (LLMs). By submitting malicious prompts, attackers could potentially manipulate Copilot into exposing confidential organizational data that would normally remain protected. Because Microsoft 365 Copilot integrates deeply with enterprise environments, the risks include unauthorized access to emails, Teams conversations, Word documents, SharePoint files, financial records, and other sensitive corporate information. The vulnerabilities are tracked as CVE-2026-26129, CVE-2026-26164, and CVE-2026-33111, each assigned a CVSS 3.1 score of 7.5. The first flaw involves improper neutralization of special elements, while the second stems from injection issues affecting downstream processing components. The third vulnerability impacts Copilot Chat in Microsoft Edge and could allow command injection within the browser chat environment. All three vulnerabilities primarily affect data confidentiality rather than system integrity or availability. Microsoft confirmed that the vulnerabilities have already been mitigated through backend security updates and enhanced input validation controls deployed across its managed cloud infrastructure. As the fixes were implemented centrally, organizations using Microsoft 365 Copilot and Edge Copilot Chat are automatically protected, and no additional customer action is required at this time.