Frontier AI Models Are Shockingly Gullible | Snehal Antani artwork

Frontier AI Models Are Shockingly Gullible | Snehal Antani

MTS

November 13, 2026

Horizon3 co-founder and CEO Snehal Antani discusses the launch of the company’s Adversarial Intelligence Frontier Lab and how AI agents are changing offensive and defensive cybersecurity.

0:00

Snehal Antani

!,In In our our Adversarial Adversarial Intelligence Intelligence Lab Lab, we we did did research research on on the the gullibility gullibility of of AI AI models models. So So it it turns turns out out that that the the Frontier Frontier Labs Labs, the, the OpenAIs OpenAIs, the, the Anthropics Anthropics, the, the Chinese Chinese open-source open-source models models and and others others, they they all all trained trained on on very very shallow shallow cyber cyber training training data data, public public hack-the-box hack-the-box write-ups write-ups, public, public vulnerability vulnerability research research. That That wasn't wasn't detailed detailed enough enough for for these these models models to to become become super super effective effective in in real real world world environments environments. So So one one of of our our research research papers papers, proved proved out out that that if if you you put put a a fake fake password.txt password.txt file file in in a a cyber cyber range range, a a human human hacker hacker will will get get tricked tricked by by that that fake fake file file 37 37 percent percent of of the the time time.

An An AI AI agent agent will will get get tricked tricked by by that that same same file file almost almost 96 96 percent percent of of the the time time.

0:44

SPEAKER_2

Hello Hello, everyone everyone and, and welcome welcome back back to to MTS MTS. Today. Today, we we are are joined joined by by Snehal Snehal Antani Antani, the, the co-founder co-founder and and CEO CEO of of Horizon3 Horizon3. It's It's Node Node Zero Zero platform platform has, has companies company's own own networks networks before before attackers attackers do do. And Today today, Horizon3 Horizon3 is is launching launching its its Adversarial Adversarial Intelligence Intelligence Frontier Frontier Lab Lab. So. So Snehal, Snehal, welcome welcome to to MTS MTS.

1:05

Snehal Antani

Well, I'm I'm glad glad to to be be here here. Thanks, thanks for for the the time time.

1:07

SPEAKER_2

Thank Thank you you for for the the time time, I'm. I'm really really excited excited to to have have you you on on, and, and especially especially because because I I want want to to talk talk about about what what an an Adversarial Adversarial Intelligence Intelligence Frontier Frontier Lab Lab is is. So. So what what did did you you guys guys announce announce today today?

1:18

Snehal Antani

So So we we pioneered pioneered at at Horizon3 Horizon3 this this idea idea of of AI AI hackers hackers. In In fact fact, we, we had had a a nine-year-old nine-year-old kid kid use use our our product product a a couple couple of years years ago ago to to hack hack a a bank bank in in four four minutes minutes and and 12 12 seconds seconds.

We We keynoted keynoted BlackHat BlackHat with with the the NSA NSA on on how how we we hack hack an an organization organization in in 77 77 seconds seconds, which which means means if if your your defenses defenses can't can't stop stop us us in in 76 76 seconds seconds, it's, it's game game over over. So So the the future future of of AI AI, or or future future of of cyber cyber warfare warfare is is AI AI versus versus AI AI, with with humans humans by by exception exception.

1:42

SPEAKER_2

We.

1:42

Snehal Antani

We kind of invented invented this this space space. So So the the Adversarial Adversarial Intelligence Intelligence Lab Lab is is all all about about now now taking taking all all of of this this training training data data we've we've collected collected from from all all of of our our customer customer environments environments, on on how how we've we've hacked hacked them them, how, how their their defenses defenses worked worked or or didn't didn't work work, and and building building out out the the next next generation generation of of attack attack and and defense defense models models in in order order to to help help our our customers customers fix fix things things even even faster faster and and become become more more resilient resilient to to real real cyber cyber attacks attacks. And. If if you you see see what's what's happening happening in in the the news news today today, attackers attackers are are only only getting getting faster faster, and. And so so I I think think Adversarial Adversarial Intelligence Intelligence has has never never been been more more important important.

2:12

SPEAKER_2

Yeah Yeah. I, I wanted wanted to to ask ask you you about about, because because this this is is everything everything that's that's happening happening in in the the news news right right now now. You're. You're hearing hearing about about cyber cyber attacks attacks, you're, you're hearing hearing about about hacking hacking. What What is is unique unique in in 2026 2026 around around the the general general landscape landscape for for how how these these hacks hacks happen happen?

Read the other 26 minutes of this episode here.

$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.

Not what you expected? Full refund within 14 days.

Thousands of transcripts fetched by people building searchable podcast archives

Fetch transcripts with code

The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:

request
curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.

Using your own key:

request
curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000801652145