Summary
In this a16z conversation, Armadin’s Kevin discusses what cyber defense looks like when attackers use AI agents. After 30 years in security and the success of Mandiant, he explains why he returned to start a new company alongside founders David Slater, Travis Lam, and Evan Peña.
Armadin uses frontier models on offense to find exploitable risk before real attackers do, under the name Armadin Red. Its second act, Armadin Blue, focuses on stopping those attacks. He says the company has found more than 90 zero-days at Fortune 500 customer sites since January, and describes coordinated swarms of around 25,000 agents working together.
The discussion covers how AI attacks compare with nation-state operations, why he thinks slowing down models won’t help because open models are already good enough, and why humans in the detect-and-respond loop will be too slow. He describes working with CrowdStrike and Palo Alto Networks so that defensive platforms can apply autonomous mitigations. He also looks at how the security operations center and the boundaries between cyber categories may change, and why CISOs should treat effective autonomous security as their goal.
📺 Source: a16z · Published October 06, 2026
🏷️ Format: Interview
