Cybersecurity Must Be Rewritten Within Years | Alex Stamos artwork

Cybersecurity Must Be Rewritten Within Years | Alex Stamos

MTS

October 26, 2026

Alex Stamos, Chief Security Officer at Cognition and former security chief at Facebook and Yahoo, discusses why the rapid rise of open-weight models marks the beginning of a dark age in cybersecurity.

0:00

Alex Stamos

We have decades of code that was written by humans, often in non-type safe languages, that cannot hold up against these superhuman bug-finding AI systems. And so we are slowly burning through all of the bugs in that software. That is what you see with these projects. And so during this period of time, it's a bit of a, like we said, dark age.

0:22

Sophia

Hello everyone, welcome back to MTS. Today, I am joined by Alex Stamos. He is the new Chief Security Officer at Cognition and the former Security Chief at Facebook and Yahoo. And today, Cognition launched its new Devin Security. Alex has a lot of interesting takes and information and why he joined Cognition. So really excited to talk to you, Alex.

0:44

Alex Stamos

Yeah, thanks for having me, Sophia.

0:45

Sophia

Thank you for joining. I think one of the places I really want to start, I want to get into all the things with Devin Security and all that stuff, but I wanted to talk about this post that you launched, the day that you actually joined Cognition. You're saying why this is important, why now? You had some really interesting perspectives here. Can you just summarize that whole post that you wrote?

1:06

Alex Stamos

Yeah. The reason I came to Cognition is it is pretty clear to me that we are entering what I've been calling, for the lack of a better word, a bit of a dark age in cybersecurity.

Most of the AI benefit up to this point is actually accrued for defenders, because we had early access to the models, we have more money collectively.

AI has been good for defense, but in just the last six months or so, what's happened is the open weight models, in particular, have gone to the level where the flaws that are being found by the cutting edge models and then fixed, can now be operationalized and turned into attacks much more quickly and with much more scale than we've ever seen before. One of the reasons I came to Cognition is I feel like I have an obligation to do something about this, and it became clear to me that really the only way I can fix these things is by working in a company that's both fixing the flaws themselves, but it's also focused on the cost effectiveness because the other thing that's become clear is the per-token cost is going to be a huge issue for security. Bad guys are running these things for free on commodity hardware. I just got my M5 Studio Ultra at home. I'm running obliterated GLM53 to do experiments on, and it's amazing what you can do with your home consumer hardware. That cost effectively nothing per token, whereas if you're paying mythos level prices, you're not going to survive against a bunch of attackers. That's one of the things I really appreciate about Cognition, is we're actually constantly cutting prices and trying to work to build our own models that are much cheaper than the frontier models, but can do the same kind of security work.

2:43

Sophia

Yeah. This is particularly interesting. This past week, I would say cybersecurity went mainstream. I feel like, of course, we've been hearing about it for the last few months, but this past week was the first time I feel like the everyday user was, oh my gosh, I'm starting to understand the D-day of how this actually will affect our internet. You have this really interesting line that you wrote as well. You say, it's become very clear to me that the core of the cybersecurity problem over the next several years will not just be technological, but economic. And this piece around economic, which you just alluded to, I feel like is this missing area that people aren't talking about enough. Can you just unpack that a bit more?

3:20

Alex Stamos

Yeah. So, you know, our friends at the big labs at the Open AI and Anthropic, they've done a great job of building these Cybertune models. Mythos is the most famous, but Open AI has some fantastic ones too. And in the new Devin Security we launched today, you have access to those actually, right? So we believe that this is probably the broadest access we, anybody's been able to offer Claude Mythos to basically all of our, pretty much all of our enterprise customers, we can now offer it. So instead of like a tiny little list of companies, a huge number of companies are now going to be able to scan their code with Mythos. That's fantastic. It's a really good model. It's spectacularly expensive. We're giving away scans for free right now. So we've also announced that we've spent hundreds of millions of dollars getting free scans to enterprise customers, because we do feel like there's obligation. But we can't afford that forever, and that is on the remediation side. You can't use models at expensive if you want to fix all the things we need to fix. We have hundreds of billions of lines of code out there. It's not just everybody thinks about Linux and Windows and the things we all share. But if you think about every enterprise, some of our financial customers have tens of thousands of GitHub repositories hooked up to Devin.

Read the other 14 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/1000798329670