AIThe Hacker Newsabout 3 hours ago
Microsoft Warns Poisoned MCP Tool Descriptions Can Make AI Agents Leak Data
1 min read
Microsoft research shows attackers can hijack AI agents by poisoning MCP tool descriptions. The agent follows instructions in plain text, never breaking rules, so no alarm fires. The attack targets agents that act on user behalf, like Microsoft 365 Copilot, which can send email and change calendars.
Level
Hype check
Tap to vote and see what everyone thinks.
In this storyMicrosoft
