Blog
How AI agents get attacked, how they get tested, and how source code stays secure — from the team building RingZero.
Tool poisoning hides instructions inside MCP tool descriptions where users never look but agents always read. Here's how the attack works and how to detect it.
Read the post →
A practical nine-point checklist for reviewing any MCP server — package or remote endpoint — before it gets your agent's privileges.
Read the post →
AI pentest agents reason about targets instead of matching patterns. What that means in practice, how a run is structured, and why authorization is the hard part.
Read the post →
npm spent ten years teaching the industry how package registries get attacked. MCP registries are about to take the same course — with agents amplifying the blast radius.
Read the post →