Blog

Securing AI agents, in the open.

Zero-Trust fundamentals, prompt injection, MCP server verification, and walkthroughs of real attacks from our red-team corpus.

RSS feed

Get new posts by email. No spam.

Latest · June 13, 2026

Attack explained: the credential exfiltration kill chain

A walkthrough of one of the most common agent attacks, reading a secret and sending it somewhere it should not go, and why single-call checks miss it.

Read the post

June 11, 2026

Verify an MCP server before you connect to it

An MCP server you connect to can read your context and propose tool calls. Check its card, signature, and supply-chain risk first, the same way you would vet a dependency.

Read

June 10, 2026

Prompt injection is a tool-call problem

Prompt injection only matters when it makes an agent do something. Catching it at the tool call, not the prompt, is what actually stops the harm.

Read

June 9, 2026

What Zero-Trust means for AI agents

Agents act on your behalf with real credentials and real tools. Zero-Trust for agents means never trusting a tool call by default, and verifying every one before it runs.

Read