This transcription is provided by artificial intelligence. We believe in technology but understand that even the smartest robots can sometimes get speech recognition wrong.

 Everyone. I'm Drex and this is the two minute drill. Thanks for being with me today. Here's some stuff you might wanna know about. There's a number in cybersecurity called Time to Exploit, TTE, and it's, it measures the gap between when a vulnerability is announced, usually that includes a patch, and when attackers weaponize the vulnerability, that gap is what defenders have to work with.

They ideally need to patch. That vulnerability before it can be weaponized. In 2018, that window was about 63 days. By 2022, it was 32 days. By 2023, Mandiant measured it at about five, and then it went negative, meaning exploitation. Was happening before a patch existed, before your team had anything to do, anything to act on.

As of last year, nearly 29% of exploited vulnerabilities were weaponized on or before the day a patch was released. Which brings me to CVEs. This is a little technical kind of bear with me. It helps the rest of the story play out Common vulnerabilities and exposures, CVEs. Think of CVEs as the universal diagnosis code for software flaws.

A researcher finds a problem and then the software owner creates a patch, and then the vulnerability gets a CVE number, and that's what everybody refers to. That's what triggers your alerts and your scanners and your patch. Notice your whole patch program. Without a CVE, you're blind to the vulnerability time to exploit TTE.

That only starts counting from the moment A CVE exists, and that's the assumption the entire patching plan is built on. But stay with me because something happened last week that may have broken that whole model entirely. Now let's talk about a software product called Ghost. Ghost is a content management system, the kind that lots of organizations use for websites and newsletters and public announcements.

It's been around for 20 years, and in the last 20 years it's never had a critical security vulnerability, not even one. Well, last month, anthropic researcher Nicholas Carlini, pointed an AI model called Claude Mythos at the Ghost Code Base. Mythos is the most advanced model Anthropic has ever built. It's not released yet, one that they've been privately briefing the US government about.

Warning that because of the way it works, it makes large scale cyber attacks significantly more likely. So the researcher points mythos a ghost, and he tells it to see if it can find a vulnerability. And in under two hours, mythos found a vulnerability. Ghost didn't know was there. And then without any additional human direction, it wrote its own exploit and navigated the flaw on its own and extracted admin, API, keys and password hashes, all without authentication, it found a vulnerability no one else had ever found, and then it weaponized that vulnerability and 90 minutes.

By the way, that's now a CVE rated as highly critical, and it's a problem with 15 plus versions of Ghost. We only know what we know about Mythos at this point because Anthropic leaked the information on accident. A configuration error in their own content management system left thousands of internal documents public.

Including a draft describing mythos capabilities. Days later, anthropic accidentally leaked again this time, 500,000 lines of Claude code source code now on the public internet. Two self-inflicted disclosures in one week, all from the company that at the same time was warning Washington about how dangerous this new version was going to be from a cybersecurity perspective.

Now let me connect the dots. The window from vulnerability announcement to weaponization has gone from 63 days to 32, to five to negative, and now there's an AI tool that doesn't even wait for a known vulnerability. It finds new ones and weaponizes them in the same afternoon. Your attack surface as a healthcare organization, all those things you use every day.

EHR patient Portal, CMS platforms. Vendor integrations, medical devices, all those hundreds of little apps across your network. Many of those haven't had a critical CBE in years, and that used to kind of make some executives feel safe. But what it actually means, what we know it means now, is that you're likely to have an undiscovered vulnerability that nobody has looked hard enough to find yet.

Your current patch program was designed for human researchers who need days or weeks to reverse engineer a patch and then weaponize it. AI needs an afternoon to discover a whole new vulnerability and then weaponize it. The adversaries who get access to tools like this aren't really planning their attack around your patch cycle, so we'll now have to refocus on continuous vulnerability discovery over waiting on vendors to release patches.

Identity and network segmentation become even more important. And patch cycles need to be measured in days or hours and not weeks. Vendor conversations have to go beyond, have you had a breach to how fast can you find and fix something new and board conversations will have to evolve to treat AI powered attack speed as today's reality.

Today's baseline, not tomorrow's risk. A lot of that stuff we've been paranoid about. Some of it's already here and we just watched it work. Thanks for being here. That's it for today's two minute drill. You can find all our security podcast, including the ones you might have missed at this week.

health.com/unh. Hack. Thanks for being here. Stay a little paranoid. I'll see you around campus.