Microsoft Copilot just exfiltrated a company's files. The attack was one email. Here's the mechanism.

Microsoft Copilot just got tricked into exfiltrating a company's files by a single email — and the bug isn't really Microsoft's. It's the default behavior of every LLM-with-tools shipped in 2026. Here's the exact attack chain, plus the four mitigations that actually cut the risk in your own AI features.

Read Original

Related

Dev.to tutorial 35m ago

Stop the action, not the injection

2026's agent attacks are one bug. The fix isn't a smarter model — it's a signed limit at the tool call, outside the agent. Four runnable demos.