Feb 4, 2026 · 38m · american-optimist
AI Expert: Agents Will Take Over Everything · Joe Lonsdale
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of American Optimist, host Joe Lonsdale interviews cybersecurity pioneer Kevin Mandia to explore the history of state-sponsored cyber espionage and the urgent security threats posed by autonomous AI agents. Mandia outlines the evolution of cyber warfare from early Air Force investigations to modern enterprise defense, emphasizing why continuous AI-driven red teaming is essential for future protection.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Joe holds 12.6% of the talking time here. How this is scored →
Joe as informed peer 3.0 Guest teaching 2.9 Guest disagreement 0.4 Joe pushing back 0.0
speaking balance: gold is Joe, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 21:58 Mandia rejects the 'you sold your company' premise
Mandia directly corrects Joe's phrasing, clarifying that public companies are perpetually on the market rather than founders actively deciding to sell.
Hardest push from Joe ▶ 27:17 Joe presses on open-model weaponization risksJoe challenges the idea that state actors can keep capabilities caged by pointing out how actors in Eastern Europe iterate on open models.
Biggest teaching moment ▶ 12:05 Mandia dismantles the assumption that corporations can defeat nation statesMandia recounts confronting a financial services client to explain that expecting private enterprises to pitch a perfect defensive game against nation states is fundamentally unrealistic.
Joe holds their own ▶ 5:13 Joe cites PayPal frontline experience with Russian syndicatesJoe injects firsthand domain knowledge regarding Russian financial fraud operations during his time at PayPal, which Mandia immediately validates.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Joe as informed peer | Guest teaching | Guest disagreement | Joe pushing back | Why |
|---|---|---|---|---|---|---|
| Show Title Sequence and Host Background Montage | 2 | 3 | 0 | 0 | Joe introduces Mandia and connects early internet history to his own 1993 AOL experience. Mandia educates on early classified military intrusions, Cliff Stoll, and the Morris worm. | |
| Decades of Combat: State Actors and Cyber Crime | 4 | 3 | 0 | 0 | Joe brings relevant domain experience from his early days at PayPal observing Russian cyber syndicates. Mandia validates Joe's observation by naming specific Russian actors and detailing weapons software espionage. | |
| The Founding of Mandiant & Corporate Breach Realities | 2 | 4 | 1 | 0 | Joe prompts Mandia on founding Mandiant and its acquisition. Mandia explains why traditional antivirus is ineffective and details the harsh reality that private enterprises cannot independently stop targeted nation-state attacks. | |
| High-Stakes Cyber Investigations and Ransomware Crisis Management | 2 | 3 | 0 | 0 | Joe asks about operational war stories and law enforcement deterrence. Mandia explains shifting Russian tradecraft and why foreign state-sponsored criminals remain untouchable by Western law enforcement. | |
| Scaling Cybersecurity Talent and Re-Entering the Arena with AI | 3 | 4 | 2 | 0 | Mandia gently corrects Joe's framing that he chose to sell his company, explaining public company buyout dynamics. He details how AI will scale elite brain-surgeon-level cyber talent into swarms. | |
| Offensive AI Agents, Social Engineering, and Asymmetric Threats | 4 | 2 | 0 | 0 | Joe articulates corporate vulnerability concerns looking forward to 2026. Mandia builds on this with examples of autonomous voice cloning and asymmetric offensive cost advantages. | |
| Protecting Critical Infrastructure and Boardroom Red Teaming | 5 | 3 | 0 | 0 | Joe exhibits solid technical knowledge regarding safety guardrails and jailbreaking differences across Claude, xAI, and open-source models. Mandia explains why realistic red-teaming beats compliance dashboards in boardrooms. | |
| The Case for AI Optimism and Program Conclusion | 2 | 1 | 0 | 0 | Joe invites an optimistic closing argument on AI security. Mandia outlines an analogy comparing offensive AI training defensive AI to Tom Brady training a defense. |
Statements from this episode (13)
Mandia: US Military Began Monitoring Its Networks Around 1993
“I started doing computer security in 1993, and that was about the year we started putting eyes on our network in the military.”
Mandia: Mandiant and colleagues have responded to Chinese intrusions daily since 1995
“I think by 1995, until now, I would say on a daily basis, either I'm
Or someone I work with has been responding to an intrusion from a Chinese actor looking for the government.”
Mandia: China matched Russia's elite cyber intrusion capabilities by 2020.
“And I do believe you know, Russia had the crown early, 19 nineties, early 2000, the best intrusions I saw.
Was Russian operators on our networks.
later in my career, especially around 2019, 20, 20 China, you know, really rapidly graduated to great capabilitie…”
Mandia: Non-developers can bypass legacy antivirus software in under three minutes.
“We could teach FBI agents in the classroom who were not malware developers how to circumvent antivirus in under three minutes. Compress and encrypt malware. And you got past antivirus.”
Mandia: Mandiant originally aimed to build the endpoint software CrowdStrike perfected.
“Believe it or not, Mandiant was actually an endpoint company. I just executed so poorly on that dream. Nobody knows it, you know, but we were building an endpoint to detect what Symantec and McAfee missed... And since that time George Kurtz and CrowdStrike kin…”
Mandia: Nation-states will always beat private companies in cyberattacks
“When a nation state targets a company, guess who wins? The nation does, not the company... You cannot expect great blue chip companies that constantly every day pitch a perfect game on defense against nations that are targeting them. It's just an unfair fight.”
Mandia: Russian SVR doctrine shifted in August 2015 toward data leaks
“I had 20 years of responding to SVR. They never hacked and then released data. They suddenly did that in August, They always hacked for security reasons, and all of a sudden we're responding to them hacking universities. I'm like, oh, they increased scope. And…”
Mandia: Uniformed North Korean government agents run cyberattacks to steal Bitcoin.
“Literally government agents in North Korea in uniform are hacking to steal Bitcoin.”
Mandia: Cybercrime and Espionage Will Shift from Human-Led to AI Agent-Led
“You're gonna have AI agents on offense, automating incredibly talented humans, you know, like literally that brain surgeon analogy. AI agents will be that, and you're gonna have swarms of agents that can communicate, and think, and instantiate normal, and lear…”
Mandia: Fully autonomous cyber defense systems will emerge within three years.
“Fast forward two or three years, Joe, you're going to see an autonomous defense crafted, trained by the world's best cyber offense, you know, the hyper attack platform.”
Mandia: Good guys will use offensive AI agents to train defensive AI
“There is no question that in the future you'll have AI agents on offense run by the good guys to train the AI on defense run by the good guys.”
Mandia: AI-native software developers are over 100 times more productive.
“It's not 10 X. It's more than a hundred X productivity with an AI native developer working on software.”
Mandia: AI attackers will eliminate the 10-minute human incident-response window.
“The day where a human would break in, and if you were 10 minutes behind the human on defense, you could stop the impact of a breach, is going to end. There's going to be an agent that breaks in, then another agent uploaded because it had remote code access, Re…”