Writing

Blog

Security research and technical writing on AI agent failure modes, the AFB taxonomy, and the infrastructure for keeping agents under human control.

What is an Agent Failure Boundary?

AFB01 through AFB04 define the four distinct ways an AI agent system can fail — at the context input, at the model boundary, at the model output, and at execution. Here is what each one means and why the distinction matters for anyone building agents in production.

Read more →