The episode centres on cybersecurity executive Kevin Mandia’s view that advanced AI is becoming a decisive tool in cyber conflict and will compress the time needed to discover and exploit vulnerabilities. Mandia describes building a new company focused on automating red-team style attacks and pairing experienced penetration testers with “AI native developers”. He claims work that previously took several days can now be completed in minutes, and argues AI-enabled attackers will be able to probe targets at scale by running vast numbers of parallel attempts rather than finding a single path in.
Mandia discusses the proliferation of sensors and data — including driver-facing cameras in vehicles — arguing that while systems may become harder to compromise as code improves, the consequences of a successful breach could be more severe given society’s dependence on connected devices and networks. He suggests a shift towards “bespoke” security models trained on an individual’s normal behaviour, potentially prompting for human approval when unusual actions occur.
On cyber warfare, Mandia predicts specialised, vertical AI models optimised for attacking particular targets (such as phones, drones, or operating systems). He recounts a 2000-era case involving Russian cybercriminals who automated online fraud, presenting it as an example of how criminals rapidly adopt new efficiencies. Mandia argues defence will require building offensive capability to test and harden systems, including AI-driven tools that can iterate quickly, potentially even self-improve, to match adversaries’ pace.
Cybersecurity executive Kevin Mandia says his firm’s red-team work has been accelerated dramatically by AI, claiming tests that previously took “5 days” can now be done in “5 minutes to 10 minutes”. He presents this as already operational capability, and argues the same speed-up will be available to criminal and state actors as AI takes “the human out of the loop” in intrusion work.
Read this moment →Kevin Mandia argues that cyber defence will require building offensive AI systems capable of iterating on themselves, describing a goal of creating an “AI attacker” that can build its “next AI attacker”. He frames this as a necessary response to an AI-driven escalation in cyber operations, contending that if Western firms and governments do not develop such capability, adversaries will.
Read this moment →