AI tool poisoning exposes a major flaw in enterprise agent security

AI agents choose tools from shared registries by matching natural-language descriptions. But no human is verifying whether those descriptions are true.

Raccontata daventurebeat.com