Sandbox the AI agent. Or the next incident is already in your repo.
The single highest-leverage control most teams running AI dev tools are missing. Why sandboxing is the meta-control after the Cursor/Railway incident.
Read article ↗I threat modeled a customer support agent. Here's what I found.
A walkthrough of a real agent assessment using the ATLAS framework, including the exfiltration path nobody saw coming.
Read article ↗The 7 ways AI agents get compromised.
A practical attack taxonomy for AI agents, with concrete scenarios for each one. Not theoretical, not hand-waving.
Read article ↗The tool combination problem nobody's auditing.
Individual tools look safe. Combined, they create exfiltration paths, fraud chains, and privilege escalation.
Read article ↗System prompts are not guardrails.
Why "don't share customer data" in a system prompt is a suggestion, not a security control, and what to use instead.
Read article ↗Shadow agents: the AI security problem you can't see.
Your employees are running AI agents with personal API keys and company data. How to find them and what to do about it.
Read article ↗5 questions every CISO should ask about AI agents.
A quick diagnostic for security leaders to assess whether their organization's AI agents are governed, monitored, and controlled.
Read article ↗