Episode overview

The New Apex Predator: CrowdStrike CEO George Kurtz on Adversaries Armed With AI Agents

No Priors Podcast · 45m · 5 Egleze moments

What happened in this episode?

CrowdStrike founder and CEO George Kurtz tells the No Priors podcast that the spread of AI agents inside enterprises is outpacing organisations’ ability to track them, describing ‘shadow AI’ as a growing governance and identity problem. He argues that AI is also compressing the traditional hierarchy of cyber adversaries, claiming ‘ablated’ and open-weight models without guardrails can give low-sophistication actors capabilities that previously resembled nation-state tradecraft.

Kurtz describes operational changes he says CrowdStrike is observing, including sharply reduced ‘breakout time’—the interval between initial access and lateral movement—claiming an average of 29 minutes and cases as fast as 27 seconds. He also contends that defenders increasingly face prompt-driven attacks that can generate unique behaviours, making signature-based detection harder.

On the defensive side, Kurtz outlines CrowdStrike’s ‘SafeMind’ initiative, including red- and blue-team models and an approach he says was influenced by Nvidia’s push for ‘digital twins’ of customer environments. He argues that frontier AI models’ safety refusals can hinder defensive work, and claims in-house security models can reduce per-task costs dramatically, while also addressing customers’ concerns about sending sensitive data to external AI labs.

Kurtz predicts enterprises will rely more on open-weight models as token costs rise, and he is sceptical of claims that society will need to ‘turn the internet off’, arguing agents generally replicate human techniques faster rather than inventing entirely new classes of attack.

What are the key points?

Kurtz says ‘shadow AI’ is making it difficult for firms to know where agents are running and under what identities.
He argues AI agents flatten the ‘pyramid’ of adversaries, with non-state actors gaining more sophisticated capabilities.
Kurtz claims CrowdStrike’s tracked intrusion ‘breakout time’ has fallen to minutes on average, with extreme cases measured in seconds.
He says attacks are shifting towards prompt-driven behaviour that can produce unique ‘signatures’ per intrusion path.
Kurtz describes CrowdStrike’s ‘SafeMind’ effort, including red/blue security models and the use of ‘digital twins’ to iterate defences.
He predicts more enterprise adoption of open-weight models due to cost pressures, while urging defender access to frontier-calibre capabilities.

Who is speaking?

Named speakers identified in Egleze's source-linked moments from this episode.

George Kurtz

Which topics are covered?

AI & Tech Crime & Justice Money

How is this page sourced?

Programme: No Priors Podcast
Duration: 45m
Original episode: Open the original source ↗

Egleze verifies the speaker, source and timestamp for each linked moment. Unless a moment is explicitly marked independently verified, statements remain attributed claims by the named speaker rather than established facts.

Watch original episode More from No Priors Podcast

5 moments from this episode

Source-linked · editorially selected

AI & Tech

CrowdStrike CEO George Kurtz says AI agents are flattening attacker hierarchies

George Kurtz · at 3:18 · Attributed speaker claim

Well, that has been totally just blown up because really now what we're seeing is the new apex predator is the agent state, not the nation state, it's the agent state. And what the agent state has done, if we think about the the models that are out there um that are bladed, so there there's no real guard rails. It's essentially taken all of these very complex topics and sort of techniques and cost and now flatten the entire pyramid.

CrowdStrike CEO George Kurtz argues that AI-driven agents and models with few guardrails are shrinking the gap between nation-state hackers and less sophisticated actors. He says this has flattened the traditional adversary pyramid by speeding up vulnerability discovery and attacks.

Read this moment → Check original at 3:18 ↗

Crime & Justice

CrowdStrike CEO Kurtz warns open-weight models run locally can lower barriers to building ransomware

George Kurtz · at 5:44 · Attributed speaker claim

I've got two Spark DGXs at home and you know I've got all the models running and you know whatever I want to do I can do. You want to create ransomware no problem. You want to create a ransomware kit, no issue.

George Kurtz says attackers are often turning to open-weight models because frontier models have stronger guardrails around cyber-related requests. He claims that locally run, less-restricted models—ones he says can be operated on powerful home hardware and increasingly capable personal machines—make it easier to generate or iterate on offensive tooling such as ransomware and “ransomware kits”.

Read this moment → Check original at 5:44 ↗

Crime & Justice

CrowdStrike CEO Kurtz says breakout times average 29 minutes

George Kurtz · at 7:44 · Attributed speaker claim

So on a that used to be days and you know hours and now it's minutes. On average it's 29 minutes but we've seen it as quick as 27 seconds.

George Kurtz says CrowdStrike tracks “breakout time” as how long it takes an intruder to pivot laterally after first landing on a machine. He claims the average is 29 minutes and that some cases have been as fast as 27 seconds.

Read this moment → Check original at 7:44 ↗

AI & Tech

CrowdStrike CEO Kurtz says cyberattacks are shifting from ‘malware’ to prompt-driven tactics

George Kurtz · at 8:01 · Attributed speaker claim

And what people have to keep in mind is it's not malware anymore. It's prompts. So they can basically engineer prompts that would uniquely create create a totally unique signature of what direction they're going to go in, how they gather information, where they pivot.

George Kurtz argues that some adversaries are increasingly using prompts to drive actions and reconnaissance, producing highly variable behaviour that may not match reusable malware signatures. He says this rapid iteration makes it harder to distinguish and track attacks using traditional signature-based approaches.

Read this moment → Check original at 8:01 ↗

Money

CrowdStrike CEO Kurtz says in-house security models can cut per-task costs versus frontier cyber models

George Kurtz · at 20:53 · Attributed speaker claim

So for general remediation if you use like uh the highest caliber frontier cyber model it probably cost you about $10. Ours is like 3 cents.

George Kurtz says CrowdStrike’s ‘SafeMind’ effort includes models post-trained on the firm’s security data, and argues this focus can deliver useful security outputs at lower cost. He offers an example that a “highest calibre frontier cyber model” might cost about $10 for general remediation, while CrowdStrike’s model costs about $0.03. He also says some customers are concerned about sending sensitive data to external AI labs or having it used for training.

Read this moment → Check original at 20:53 ↗