The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents
George Kurtz argues that AI agents have flattened the adversary pyramid and made attacks faster, but defenders can win by fighting AI with better AI, built on their own data, digital twins and open-weight models.
Watch on YouTube ↗Charlie's summary
- ▶ 3:01
The new apex predator is the agent state: AI has flattened the adversary pyramid, giving activists nation-state capability
George describes a pyramid of pain with activists at the base, e-crime in the middle and nation states on top. Unguarded models have collapsed it, so the capability a nation state had a few years ago is now in an activist's hands. The result is more vulnerabilities found and attack after attack. An agent with enough time “needs to get in or it dies.”
- ▶ 5:03
Attackers now use prompts, not malware, and breakout time has shrunk to minutes
George says breakout time, how long an intruder needs to pivot from the first machine, used to be days or hours. It now averages 29 minutes and has been seen at 27 seconds, set by the speed of inference. “It's not malware anymore. It's prompts,” so each attack can have a unique signature. Attackers lean on open-weight or ablated models because frontier guardrails block cyber work. George runs such models on two DGX Spark machines at home and says they will write ransomware on request.
- ▶ 9:06
Agents get your identity and download anything, so shadow AI is the new seam adversaries exploit
George says adversaries look for gaps in new technology. Companies once vetted every package before running it. Now they give an agent their own identity, or worse, and tell it to download whatever it wants, such as npm packages. Nobody watches or understands what it does. Organizations first need to know where their agents are and what identity each runs under. CrowdStrike can trace an agent's whole chain of sub-agents and tool calls and add policy and blocking rules to it.
- ▶ 11:40
The Hugging Face incident showed defenders lacked the AI that attackers had
George's main takeaway is not the first end-to-end agentic attack. It is that defenders lacked equal capability. Hugging Face asked frontier models whether the incident was real and was refused, so it had to turn to open-weight models to sift the data. Defenders need frontier-grade defensive AI that does not refuse cyber work. That is why CrowdStrike built Safe Mind, with a red model (Tempest) and a blue model (Solano).
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.