Zero-click AI data leak flaw uncovered in Microsoft 365 Copilot

Researchers at Aim Labs discovered a zero-click vulnerability, 'EchoLeak,' in Microsoft 365 Copilot that allowed for the exfiltration of sensitive data without user interaction by using a malicious email with a hidden prompt injection. Microsoft classified the information disclosure flaw as critical (CVE-2025-32711) and has since patched it server-side, with no evidence of real-world exploitation affecting customers. This attack highlights a new class of vulnerabilities, 'LLM Scope Violation,' and underscores the need for enhanced security measures, including stronger prompt injection filters and granular input scoping, as AI integration deepens in business workflows.

Analysis

Researchers at Aim Labs identified a significant zero-click vulnerability, 'EchoLeak' (CVE-2025-32711), within Microsoft 365 Copilot, which could have enabled attackers to exfiltrate sensitive data without user interaction. Microsoft classified this information disclosure flaw as critical and implemented a server-side fix in May 2025, confirming no evidence of real-world exploitation or customer impact. The attack leveraged a malicious email with a hidden prompt injection designed to bypass Microsoft's XPIA classifiers, which, when processed by the Retrieval-Augmented Generation (RAG) engine, could trick the LLM into leaking internal data via trusted Microsoft Teams or SharePoint URLs. While this specific vulnerability is remediated, EchoLeak's significance lies in demonstrating a new class of vulnerabilities termed 'LLM Scope Violation,' highlighting the emerging risks associated with the deep integration of large language models into enterprise workflows. The incident, reflected by a mixed sentiment (-0.1) and cautious tone, underscores that increasing AI complexity may outpace traditional security defenses, necessitating enhanced measures such as stronger prompt injection filters, granular input scoping, and post-processing filters on LLM outputs. The low market impact score (0.3) suggests the immediate financial effect on Microsoft is limited due to the effective patching and lack of exploitation, but the event serves as a cautionary signal for the AI sector regarding novel attack vectors.

Key Decisions for Investors

• Investors should monitor Microsoft's ongoing R&D and disclosures regarding AI security enhancements, particularly for LLM-specific vulnerabilities, as the 'EchoLeak' incident signifies a new category of risk for AI-driven products.
• Consider the potential for increased cybersecurity expenditure across the AI sector, including for Microsoft (MSFT), as companies fortify defenses against sophisticated threats like LLM Scope Violations, which could impact profit margins.
• While Microsoft's swift remediation of CVE-2025-32711 mitigated direct impact, this event highlights the evolving threat landscape; thus, evaluating the cybersecurity posture and resilience of AI-integrated systems should be a key due diligence point for investments in this space.