AgentsJuly 29, 2026via Simon Willison
AI Worming through Word
Why it matters
AI agents integrated into productivity tools create a new attack surface. Malicious documents can now exploit agent capabilities to propagate across enterprise networks—a security category that barely existed 18 months ago.
Key signals
- Attack vector: AI agents embedded in Microsoft Word exploited via document macros
- Threat model: lateral movement and propagation through workplace documents
- Timeline: July 29, 2026—agent security vulnerabilities now weaponized in the wild
- Implication: enterprises running agentic AI in mainstream productivity tools face novel infection paths
The hook
AI worms are now spreading through Microsoft Word. Here's how attackers are weaponizing document macros to turn AI agents into a vector for lateral movement.