A hidden instruction in one Word file just proved AI assistants will infect every document they touch next.
Microsoft Copilot
+7
A hidden comment in a pull request is all it takes to hijack the AI gatekeeper your team trusts to ship safe code
Zero Trust
+6
One email is all it takes to plant a permanent lie in your AI assistant's memory — and there's no patch coming, because it isn't a bug.
identity-security
Symlink attacks, hallucinated packages, and prompt injection are turning your AI pair programmer into an attacker's easiest way in
NHI
Convention files, prompt injection, and why the line between productivity tool and data pipeline just disappeared
