Model Refusals Don't Stop Hackers | Devon Thomas
MTS
October 17, 2026
Abliteration AI CEO Devon Thomas discusses selling APIs for unredacted/abliterated cybersecurity models, red teaming AI agents, and why model refusals hinder cyber defenders. Turn ideas into software people love.
Devon Thomas
For the most part, I don't think refusals help at all for bad actors. I just don't think they're going to say, hey, Claude refused my request, I'm done trying to hack.
I don't see that, but maybe.
SPEAKER_2
And we are back, we're live with Devon Thomas, who's co-founder and CEO of Abliteration.AI, which is a company that sells an API that does cybersecurity, red teaming, with much reduced refusals compared to other products out there. So, Devon, welcome to MTS.
Devon Thomas
Hey, thanks for having me. Nice to meet you.
SPEAKER_2
Absolutely. So, you released your first product on X about, you know, hard to tell time in the singularity, but like a month ago-ish?
Devon Thomas
Well, we were, we had products, we had models before, but that was the one that everyone cared about, so we'll just go with that one.
SPEAKER_2
We'll go with that one. So, obliterated model large V2, it was a month and a half ago. And it caused like a Twitter freakout of cosmic proportions.
You said, we obliterated and hosted it, so does the offensive cyber red teaming and agent testing work? Other models refuse to do. So, you know, a lot of people's immediate response is like, this is terrible, this is like evil. Like, why would you, you know, release the, you know, a cyber nuke? This should be illegal and so on. So like, what would you say in response to those people?
Devon Thomas
Well, I'll say that, I mean, I have a relationship with a lot of people in the cyber industry. We have a ton of customers of, from huge corporations to cyber startups, and many of them were complaining about the refusals and access to essentially just be able to do their job, right? So they were so excited to get access to kind of a frontier model to help them do the work they were excited about. So, I mean, there was a lot of outsiders, I would call them like outside of the cyber industry and outside of our customer base that were worried about it. But for the most part, all of our customers were really excited to have access to these models and to better be able to defend their own products. And I think they see all the articles about all these AI hacks and etc. And they're just excited to be able to do it on themselves first before someone else with an open source model or a jailbroken frontier model does it against them. So, that's kind of, I mean, that was our goal is to proliferate access to defenders worldwide.
SPEAKER_2
Right. Well, why don't we step back a little bit and why don't you tell us more about why Abliteration was created and what the mission is?
Devon Thomas
Yeah. So, I mean, it was created with that exact purpose, right? That if you go to my, like the people I follow on Twitter, right? I follow almost all the cyber community, especially around offensive cyber security. So, I just see them complaining all day, every day about refusals and essentially, like I said, not be able to do their job, right? Because of the refusals or maybe they had access to the special cyber programs or maybe they didn't have access. Either way, it was a huge problem amongst their community that I thought that we could solve and that many companies were already kind of solving in this particular way.
And we had a, I would say, a mixed feedback, but either way, a lot of happy people that we are providing them a reliable service.
SPEAKER_2
Right. So, why do you think there hasn't been more hacking in the world already? We have very cyber capable frontier open models. Sometimes they're even obliterated.
Why hasn't there been a lot more hacking of normal people?
Devon Thomas
I think what people underestimate is how technical most cyber defenders are. And how long they've been doing this is a lot of gray hairs, a lot of gray beards, right? And they've seen all these things. And if you kind of think about models, I always say this all the time, is that models regress towards the mean.
For the most part, they're not introducing new things. And maybe they might introduce a new combination of old things, but just model training in general, right? If you think about linear regression, right? It's regressing towards the mean. So for the most part, for most cyber defenders, they're seeing things they've been seeing this whole entire time. And like maybe the, like I said, the outsiders, the people outside of the cyber community, this is new to them. But for a lot of cyber defenders, they've kind of seen much of the things that people think are being introduced by AI models.
Read the other 19 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:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown 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:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000796846836