Apr 17, 2025 · 19m · tbpn
This Company is Changing Cybersecurity | Alex Dhillon on TBPN April 10th
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 TBPN interview, Outtake AI founder Alex Dhillon discusses how his company combines Palantir-inspired engineering with autonomous agent architectures to defend enterprises and creators against the explosive rise of generative AI-powered impersonation and social engineering attacks.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The hosts as informed peer 5.2 Guest teaching 3.7 Guest disagreement 0.7 The hosts pushing back 0.5
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 9:00 Dismissing extreme AI threat hype in favor of social engineering reality
Alex pushes back against theoretical fears of models synthesizing bioweapons, countering that current models already present clear and immediate danger by democratizing social engineering and pentesting.
Hardest push from the hosts ▶ 11:53 Host interrupts to redirect toward agentic architectureThe host cuts in while Alex discusses stealth roadmap details to firmly pivot the discussion to the technical mechanics of deploying autonomous agentic bots across external web ecosystems.
Biggest teaching moment ▶ 9:01 Securing humans outside corporate firewallsAlex educates the host on modern breach statistics, highlighting that over 85% of compromises exploit human social engineering rather than internal enterprise firewalls.
The host holds their own ▶ 16:47 Host details Paul Le Roux registrar exploitationThe host showcases domain fluency by citing how master cybercriminal Paul Le Roux bypassed takedown mechanisms by establishing his own domain registrar and TLD.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Identifying Outtake AI's High-Profile Customer Base | 4 | 4 | 1 | 0 | The host asks open-ended questions about Outtake's enterprise customer base and why widespread AI impersonation Armageddon hasn't occurred yet. Alex explains the threat economics, emphasizing that attackers target enterprises for high-value data and that lowering barrier to entry lets anyone pentest. | |
| Economics of Phishing and Mirroring AI Outbound Growth | 4 | 3 | 0 | 0 | The host queries the conversion economics of phishing and creator impersonations, referencing SciComm and Human Lab. Alex draws an analogy between legitimate AI outbound SDR growth and the symmetric rise in malicious outbound social engineering. | |
| Model Progress Benchmarks and Securing the Human Perimeter | 6 | 5 | 2 | 1 | The host cites an essay on LessWrong regarding ZeroPath's benchmark saturation versus real-world autonomous hacking limitations. Alex reframes the issue, noting that 85% of critical breaches target the human perimeter rather than software vulnerabilities. | |
| Broken CAPTCHAs and Probabilistic Defense Mechanisms | 6 | 4 | 0 | 2 | The host explores probabilistic versus deterministic defenses such as Arc AGI CAPTCHAs, interrupting to steer into agentic bot architecture. Alex describes why legacy service firms failed and how context-aware agents automate seasonal scam discovery and legal remediation. | |
| Outtake AI's Value Capture and Enterprise Cost Replacement | 4 | 3 | 0 | 0 | The host asks about value capture mechanics and Outtake's operational parallels to Palantir's forward-deployed engineering model. Alex details how automated agents replace expensive round-the-clock analyst and legal teams at a tenth of the cost. | |
| Technical Mechanics and Platform Trust in Scam Takedowns | 7 | 3 | 1 | 0 | The host demonstrates deep historical context by referencing criminal mastermind Paul Le Roux's use of rogue domain registrars and asks about law enforcement escalation thresholds. Alex explains Outtake's focus on building trusted API-level reporting reputations with platforms and registrars. |
Statements from this episode (11)
Dhillon: Palantir CTO Shyam Sankar is an Outtake AI angel investor
“Worked directly for the pounder CTO, Shamsankar. Absolute goat. A shout out to him. Also one of our angel investors.”
Dhillon: Generative AI reduced the cost of malicious online behavior to zero
“The cost to be a shitty person online went to kind of zero dollars over the last, like, three years. You got infinite text, infinite images.”
Alex Dhillon: OpenAI is a customer of Outtake AI
“We've been lucky to work with phenomenal places like you know, OpenAI is one of our favorite customers really pushed us when you think about the kinds of cybersecurity threats that they might be dealing with.”
Dhillon: LLMs allow grandmothers to become cyber scammers
“It's not only that grandmothers are the ones being scammed, it's actually grandmothers can become scammers now, and it's because, right, because it's like, it's just easier than ever and that's actually the real issue. It's like, sometimes people talk about la…”
Alex Dhillon: Malicious AI outbound volume mirrors legitimate AI sales growth one-for-one
“Just as the volume of AI outbound has increased on like legitimate use cases where I'm sure you and I both deal with a ton of LinkedIn inbound or email inbound. It's basically been one for one on sort of the darker side of it as well.”
Dhillon: Creators were targeted first by AI impersonation due to public data
“The creators were the first ones to hit because they were the most chronically online, right? Is, is the short answer there. And in the early days when the models needed more data to train on, they were also the ones that had the most data available.”
Dhillon: Roughly 85% of Critical Data Breaches Stem from Social Engineering
“But I guess my claim is that, like you know it's something like 85% of critical data breaches happen because of social engineering attacks.”
Dhillon: Cybersecurity Firms Historically Hesitate to Hunt Threats Outside Corporate Perimeters
“Historically, cybersecurity companies have been kind of, I think gun shy of stepping outside their perimeter and saying, hey, no, I'm going to go search for threats. Out in the world, like arbitrary websites, social media, wherever it is, and we're going to go…”
Dhillon: AI broke CAPTCHAs and passed the Turing test without enough panic
“One of my biggest anxieties is that, like, generally, to your point, CAPTCHAs are broken, right? Which, like, hey, we passed the Turing test a while ago. For some reason, we just blew past that and don't talk about it enough. We all are blowing past what effec…”
Dhillon: External threat detection has historically been a graveyard of services firms
“Candidly, this whole space is full of a graveyard of companies that were services companies, right? So, like, it intrinsically was a task so difficult that only humans can do it.”
Dhillon: Outtake AI's edge is trusted reporting pipelines with domain registrars
“I think the big differentiator, you know, in addition to all the agentic stuff is like, we've thought really deeply about how do we become how do we become the platform that is like trusted by domain registrars, hosts social platforms, et cetera, to be a high …”