Tag: agent-security
All the articles with the tag "agent-security".
- llm-security 18 min read
Building an Agent That Can't Be Hijacked: The Design Rules
Eight frameworks tell you how to secure an agentic system and none of them are substitutes for each other. What they agree on, where they actively contradict each other, and the four properties I'm going to enforce in code in Part 2.
Read article - llm-security 26 min read
I Built the Defense, Then Attacked It: Numbers From a Secure Triage Agent
A capability-tracking interpreter for a SOC triage agent, a 20-case injection suite aimed at it, and 420 runs across three models. Exfiltration and disclosure go to zero and stay there regardless of model strength. Evasion tracks model capability, and one of my four policies turned out to be pure friction. Includes the LangGraph port, and why the framework's flagship agent pattern breaks the design.
Read article