Introduction
A newly disclosed vulnerability in Microsoft 365 Copilot BizChat could expose sensitive business data processed through the service to unauthorized parties. The issue has a CVSS score of 8.2, and it shows the risks that come with connecting AI-driven collaboration tools to enterprise data sources.
Technical information
CVE-2025-53787 is classified under CWE-77 (Improper Neutralization of Special Elements used in a Command). That classification indicates the vulnerability involves user-supplied input that is not sanitized well enough, so the underlying system could interpret it as part of a command. In Microsoft 365 Copilot BizChat, this could let an attacker craft input that is executed as a command and discloses sensitive information without authorization.
As of August 7, 2025, no public code snippets, exploitation flows, or root-cause details are available. The vulnerability's classification is similar to that of EchoLeak (CVE-2025-32711), which exploited prompt injection and LLM scope violations in Microsoft 365 Copilot. However, no direct exploitation details for CVE-2025-53787 have been published.
Affected systems and versions
- Product: Microsoft 365 Copilot BizChat
- Specific affected versions: Not disclosed in public sources as of August 7, 2025
- Vulnerable configurations: Not specified
Vendor security history
Microsoft has had several notable vulnerabilities in its AI and collaboration platforms in 2025. The EchoLeak zero-click vulnerability (CVE-2025-32711) affected Microsoft 365 Copilot and showed the risks of prompt injection and LLM scope violations. Microsoft typically ships server-side patches quickly for critical vulnerabilities, and it keeps a regular Patch Tuesday cycle. The company works with external security researchers and has a mature vulnerability disclosure process. Even so, AI-related vulnerabilities keep turning up at a pace that challenges established security models.
