AI Pentesting Found Real Kubernetes Exploits in 3 Hours for $2 Transcript, AI Summary & Key Points

00:07 A few hours of an agent's work and a few minutes of mine. That's all it took to find a way into my own Kubernetes cluster. Not they may be buried in a PDF. A real one proven with my session tokens stolen and my cluster secrets read back out to an attacker. A human doing the same thing would have taken weeks if I could even get one booked. And if you're a developer in DevOps or SR or anyone who ships software and has already decided security testing isn't your job, don't go.

00:40 Please don't go. This one's for you. It's a way to offload the work you don't want to do or never get around to and still get findings back in a shape you can act on. So here's the situation most of us are actually in. AI writes half of our code now. Actually sometimes all of it. We ship faster than we ever have. And penetration testing, you know, the thing that's supposed to catch the holes before someone else does hasn't kept up.

01:11 It's scarce. It's expensive. And it happens once a year, if it happens at all. By the time the report lands, you've shipped a dozen more releases. the version it actually tested isn't even running anymore. And half the findings are too late to fix cleanly. And what you get is usually a list of problems, not solutions. It's it's no wonder pentesters aren't most developers favorite people to hear from.

01:36 So this video is about what changes when that pentester is an AI. Now I'm going to point one at an app I actually own. let it generally break in and see what's worth. When the punest is an AI, the real challenge isn't the testing itself. It's that you're no longer rationing a scarce, expensive consultant. It runs everywhere on your laptop or a CI runner for a couple of dollars.

02:09 So, how often you run it is finally your call. Not something you can afford only once a year. Run it when you're curious. Run it on every release. Run it on a schedule. It reads your source code and attacks the running application at the same time. And here's the part that that matters really. It doesn't just list what might be wrong. It tries to prove it.

02:33 And once something's proven, an agent can pick up that report, file an issue for every finding, and start working through the fixes. Find it, prove it, fix it. All of it done by agents. Now the tool I've been trying is called Shannon from Keygraph and it's open source. It's a single command. You can run it yourself, hand it to your agent or wrap it inside the pipeline.

03:00 It's simple. And that simplicity, it's really simple. I need to stress that. And that simplicity is is a big part of what makes it good. It reads the source, attacks the running app, and reports only what it could actually exploit. Now, that last part is the whole point. Most scanners hand you a long list of potential issues, and plenty of them turn out not to be exploitable.

03:26 So, you spend real time chasing findings that were never a real risk. Shannon leaves those out. If it's in the report, there's a working exploit behind it. Here's the whole thing I run against the target. It kicks off and echoes back what it's about to test. From there, it logs in the way a user would, then works through recon, finding vulnerabilities and actually exploiting them one class at a time, injection, cross-ite scripting, authentication, authorization, and server side request forgery.

04:03 It's powered by a language model of course and it doesn't even need to be expensive one which is the whole reason this costs pennies instead of a salary of of a person right it runs in the background while I work on something else or play games maybe I don't know now every time I want to see what it's doing uh there's a status command it lays out every phase and where the time went and in this case don't freak out about the red please don't that's not something I broke the authentication agent just didn't finish on its

04:35 run. And I'll get to what that means in a minute. In a few minutes. Now, three hours later, it was done. So, that was three hours of its time and a few minutes of mine. And the whole thing cost me, I don't know, around $2, something like that. So, if you want to hire me as a pentester for your company, I will happily do it for half half of whatever the last person you hired charged you.

05:05 I'll do it half the money. Shannon did not hand me a report. Exactly. It generated a couple of them. One is a PDF which is the format you hand the manager and that person will pretend that they read it and judge it purely how big and colorful it is. That that's how managers work right now. The more pages and the charts the better and the content is really beside the point but that's for the managers.

05:32 So I will skip that one. The other one is markdown and that's the one I want because it's what I like and just as importantly what an agent likes. So the second part matters here because an agent is probably the one that's going to fix all of this. So I will open the markdown right there in the terminal just to show it to you. Now this is a long report but the summary at the top is the part that really matters and it's not a wall maybe who knows it's eight findings it actually exploited each with a working proof and

06:07 that's a lot. That's a lot for an app this small. My hobby app, right? Three of them are rated high. There is a stored cross-sight scripting bug, a static token user that inherits full cluster privileges. Imagine that. And the live session token sitting in script readable storage. Now, the one that stopped me, freaked me out a bit is the first one, the the stored cross-ite scripting bug, and it's in how the UI renders diagrams.

06:37 So one click, one click on a diagram in my app and then attacker's code runs inside my application. It reads my session token straight out of the browser and uses it to impersonate me against the cluster. Imagine that. Now Shannon did not just flag that as you know it's possible. It did it in a real browser. Then it replayed the stolen token to read back name spaces and secrets.

07:05 it should never have touched and brought it wrote the whole chain into the report. Now one note worth stating uh the demo's AI back end is a stub right so I place the malicious diagram directly instead of coaxing it out of a real model but nevertheless what matters is that everything after that the click the code running the stolen token the cluster reads Shannon did it for real none of this is magic though you know right it's it's slow it's nondeterministic so two runs won't turn up exactly the same things.

07:43 It leans hard on the model you give it. This run actually came back uh marked partial because one whole class of attack went as tested authentication failed to finish. Now that's beside the point. The findings themselves though are solid, really solid. Everyone, every single one came with a working exploit. So there is nothing to triage or second guess before an agent start just fixing them.

08:12 Here's where it gets generally useful, right? The report is the start. It's not the end. Normally, I would hand it straight to a coding agent, a different agent from the one that did the testing and let it work through the report. Open any issue for every finding and start working on each of them. I don't even come back to a report. I come back to pull requests that close the issues.

08:39 And where I used to triage down to the ones that really mattered in the past, fixing all of them, all of them now uh fall squarely into uh like why not, right? The pun tester finds stuff, the coder fixes and once that loop works one time, there is no reason to run it once a year. You schedule it. Shannon can drop into a CI pipeline. It can run on every release or just weekly.

09:09 and files what it finds automatically and from there opening those issues can trigger the fixing agent on its own. So a finding turns into a pull request without me touching anything. Security that tests itself and fixes itself release after release or on whichever cadence you set instead of an annual event you you dread really you hate. Now, could you do all this without Shannon by handing a pile of skills to a generated purpose agent?

09:46 Maybe, maybe not. Maybe after a lot of work, who knows? But that's not the real question. The real question is why would you bother? Shanner is open source and it's free. The only thing you pay is the model. So, use it. Think of it as an agent dedicated to P testing. The one that you run directly or hand to whichever agent you're already working with.

10:11 And there we go. If it earns its keep, if you say this is awesome and you want it running continuously across a team or wired into your tickets and with the dashboards and compliance a company needs, then Keyraph, the company behind it, sells a hosted platform for exactly that, right? But the part that actually does the actual work, the breaking in is the free part.

10:33 It's open source. So is this going to replace human punesters? No, it's not. And that's a wrong question. A human still catches what's this missed and brings judgment the model doesn't have. And nothing nothing here stops them. It just takes the routine sweep, the slow, expensive grind that never happens often enough. anyway and hands it to an agent for pennies.

11:02 The pun testers we have aren't replaced. They're freed, liberated for the hard critical pots. A few hours of an agent and a couple of dollars for a run. No, nobody gives up anything that matters, right? Everybody's happy. So, Shannon, it's awesome. Use it. Highly recommend it. Thank you for watching. See you in the next one. Cheers.