US EditionMicrosoft Copilot Targeted in First “Zero-Click” Attack on an AI Agent - What You Need to Know
- Researchers at Aim Labs detected a major security flaw called EchoLeak in Microsoft 365 Copilot in January 2025 that allowed data theft without user interaction.
- The vulnerability exploited a novel LLM Scope Violation technique that manipulates internal AI model logic to turn the agent against itself.
- EchoLeak permitted attackers to exfiltrate sensitive information by sending a simple email, affecting apps connected to Copilot like Word and Outlook.
- Microsoft assigned EchoLeak the critical CVE-2025-32711 identifier with a severity of 9.3/10 and fully mitigated it server-side by May 2025 without user action required.
- No evidence showed real-world exploitation, but Aim Labs warned similar attacks may increase as AI integrates deeper into enterprises, urging fundamental AI design changes.
21 Articles
21 Articles
A vulnerabilities were detected in Microsoft 365 Copilot. The failure allows an attacker to manipulate a generated IA with a simple email malicious. Without the user's knowledge, IA can send sensitive data...
EchoLeak: First-Ever Zero-Click Vulnerability, CVE-2025-3271, Discovered by Aim Labs in Microsoft 365 Copilot AI, Allowed Attackers Steal Sensitive Data Silently, Now Fixed | 📲 LatestLY
EchoLeak, the first-ever zero-click vulnerability (CVE-2025-32711), was discovered by Aim Labs in Microsoft 365 Copilot AI. It allowed attackers to silently steal sensitive user data through hidden prompts in emails without user interaction. Microsoft has fixed the security flaw with a server-side update. 📲 EchoLeak: First-Ever Zero-Click Vulnerability, CVE-2025-3271, Discovered by Aim Labs in Microsoft 365 Copilot AI, Allowed Attackers Steal S…
First ever security flaw detected in an AI agent, could allow hacker to attack user via email
Security researchers have discovered the first zero-click AI vulnerability in Microsoft 365 Copilot AI agent, exposing a way for attackers to steal data via email without user interaction. The flaw is now fixed.
Coverage Details
Bias Distribution
- 60% of the sources lean Right
To view factuality data please Upgrade to Premium
