
Sep 1, 2026 · 1h 3m
CrowdStrike bets specialized AI can outscale cyberattackers
LIVE From CrowdStrike's Fal.Con 2026 | George Kurtz, Michael Sentonas, Daniel Bernard & Justin Boitano
The episode examines whether cheaper, purpose-built defensive models can reverse the economic advantage increasingly available to AI-enabled attackers.
- 1AI makes cyberattacks more scalable, while autonomous agents create new behavior and governance problems for defenders.
- 2CrowdStrike and NVIDIA are combining threat intelligence, telemetry, open models, and specialized controls to build affordable always-on security.
- 3Customer budgets, quantum risk, social engineering, and under-resourced infrastructure complicate the push to expand advanced protection.
Don't miss
Daniel Bernard and Justin Boitano argue that specialized, low-cost defensive models could give defenders a scalable economic advantage over general-purpose offensive systems.
The brief
George Kurtz frames autonomous agents as a new security frontier, with Falcon Guardian designed to detect and respond to agent behavior as offensive capabilities become easier to access.
CrowdStrike’s SafeMind combines offensive and defensive models, deterministic controls, and real-time analysis to identify attack chains without treating AI-enabled attacks as fundamentally mysterious.
Michael Sentonas shifts the focus to deployment: customers face budget pressure, while hospitals, utilities, small businesses, and critical services still need protection against faster social engineering.
Daniel Bernard and NVIDIA’s Justin Boitano explain how open models, incident-response data, reinforcement-learning environments, and accelerated computing could make always-on defense economically viable.
The central question is whether 2026 becomes a breakthrough year for attackers or defenders—and whether specialized, lower-cost models can change that balance.
What was said on this episode
41 statements · 18 positive · 19 negative · 4 neutral
AI agents can access resources and perform harmful actions
“AI agents are more sophisticated. They have access to data. They have access to compute, to networking resources. They do bad things.”
Listen at 0:35
Security is necessary to accelerate customer AI adoption
“the missing link to accelerating AI adoption is security”
Listen at 0:59
CrowdStrike and NVIDIA created an agentic security platform
“we actually partnered with NVIDIA to create really the first, what I would call the agentic security platform”
Listen at 2:11
Compute and cost will constrain AI-enabled attacks
“the limitation for these attacks are really going to come down to sort of compute and cost”
Listen at 3:19
Agentic technology has democratized offensive cybersecurity knowledge
“the knowledge, and I talked about this in my keynote, has now been democratized”
Listen at 3:24
Agent states are now the dominant cyber threat
“the apex predator is now the agent state”
Listen at 3:43
CrowdStrike’s proprietary data gives its models a major advantage
“our models are specifically trained on our data which we think is a huge advantage”
Listen at 4:40
Customers seek maximum security outcomes at minimum cost
“customers want the best outcome at the lowest cost”
Listen at 5:00
Defenders need AI to connect indicators across attack chains
“the defenders have to use AI to be able to... connect the dots on those”
Listen at 6:49
Effective breach prevention requires deterministic security systems
“you have to have a deterministic system, which is what we built”
Listen at 8:00
AI has not invented fundamentally new hacking techniques
“It really hasn't come up with a new invention of hacking.”
Listen at 8:14
Open-weight models without guardrails can run on powerful local systems
“you can download these obliterated models and you can run them basically on a beefy system”
Listen at 9:40
Obliterated models can generate complete malware or ransomware kits
“you want a full malware ransomware kit boom it's done”
Listen at 9:51
CrowdStrike’s platform scales through one agent and control plane
“it's a very scalable model. It's a single agent. Single platform with a single control plane.”
Listen at 10:46
CrowdStrike can monitor every action performed by AI agents
“we can instrument every action that agent has taken”
Listen at 11:08
CrowdStrike’s multiple-model approach reduces AI security costs
“we're really driving down the cost”
Listen at 12:26
CrowdStrike’s defensive model continuously learns from its offensive model
“the defensive model continually learns from the offensive model”
Listen at 13:38
Large cybersecurity platform companies will win the market
“I think that's going to win in this market”
Listen at 14:16
Most Fortune 500 companies can fund AI cybersecurity needs
“they have or will find the money to deal with some of this”
Listen at 15:46
Utilities face cybersecurity disadvantages because of legacy software
“utilities, forget it. I mean, they're running such old software”
Listen at 16:05
- Michael Sentonason Organizations’ AI token budgetsNegative23:02
Many organizations exceed their AI token budgets
“every second organization has token chalk”
Listen at 23:02
- Michael Sentonason OrganizationsNegative24:14
Organizations must avoid every cybersecurity failure
“Every organization can't get it wrong.”
Listen at 24:14
- Michael Sentonason Cyber adversariesNegative25:03
Cyber adversaries disregard AI laws and privacy regulations
“The adversary does not care.”
Listen at 25:03
- Michael Sentonason AI-enabled vulnerability weaponizationNegative26:30
AI-enabled vulnerability weaponization has arrived
“We're here now.”
Listen at 26:30
Open-weight model advances have benefited cyber adversaries
“the best thing that's happened to them is the advancement in the open weight models”
Listen at 26:36
- Michael Sentonason Individual cyber attackersNegative27:17
Individual attackers at home remain constrained by compute
“The kid at home is going to be compute constrained.”
Listen at 27:17
- Michael Sentonason Nation-state cyber adversariesNegative27:20
Well-funded nation-state adversaries possess advanced AI capabilities
“if you're dealing with a nation state, if you're dealing with some of the most well-funded, well-structured, well-trained adversaries that now have this capability as well, this is the challenge”
Listen at 27:20
- Michael Sentonason Small businessesNegative28:16
Small businesses lack resources and personnel for cybersecurity
“the world's economy, I should say, runs a small business. They don't have the resource. They don't have dedicated people. They don't have the dollars.”
Listen at 28:16
- Michael Sentonason Large-scale data exfiltrationNegative30:58
Attackers exfiltrating large data volumes have an underlying motive
“there's intent”
Listen at 30:58
- Michael Sentonason AI-assisted phishingNegative32:14
AI-assisted phishing click-through rates exceed 60 percent
“Today with AI, it's over 60 now.”
Listen at 32:14
Open-weight models can provide cyberattack playbooks to inexperienced users
“you just need a model. You ask the question and it tells you what to do.”
Listen at 32:45
- Michael Sentonason AI advances and cybersecurity threatsNegative34:05
AI advances will increase vulnerabilities and cyberattacks
“We're going to get more vulnerabilities. We're going to get more attacks. It's only going to get harder.”
Listen at 34:05
- Daniel Bernardon CrowdStrike-NVIDIA cybersecurity modelsPositive44:35
CrowdStrike-NVIDIA models aim to outperform alternatives on cybersecurity tasks
“this thing needs to be better. Faster and more cost-effective than the other open source and frontier models of the day for cybersecurity use cases”
Listen at 44:35
CrowdStrike-NVIDIA harness research improved a model from 30% to 100% accuracy
“We could take a frontier model from 30% accuracy to 100% accuracy.”
Listen at 47:27
CrowdStrike will release more specialized SafeMind models and harnesses
“you'll see more models and more harnesses from us in the SafeMind family that solve different security use case problems”
Listen at 49:59
SafeMind’s harness can select frontier or open models by problem
“the harness should be able to use the best of Frontier and the best of Open”
Listen at 51:11
Using only frontier AI labs is prohibitively expensive for customers
“just going one direction with Frontier Labs is just too cost prohibitive”
Listen at 51:49
- Daniel Bernardon Cyberattack surfaceNegative53:35
The cyberattack surface has expanded substantially
“there's more attack surface than ever before”
Listen at 53:35
SafeMind shifts the balance toward cyber defenders
“SafeMind changes the curve.”
Listen at 54:10
Low-cost specialized models can provide scalable defensive coverage
“you want to help defenders have the equivalent of drones, like super low cost models that they can run everywhere”
Listen at 55:01
- Daniel Bernardon Cybersecurity marketPositive56:14
Cybersecurity demand and budgets exist for specialized solutions
“the threat's real, the need is there, the budget's there, but the market's asking for something better and something different”
Listen at 56:14
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.