Server-side request forgery

This attack has been unfolding for a while.

In May, late May, agents learned how to exploit the flaw in our artifactory, right, the SSRF flaw there.

around July 8th there was an agent called phase one one zero eight four one crazy name right um but uh decided it it wasn't going to get to where it wanted to go the way it was going so it devised a new strategy and it established a message board which is kind of what you're talking about it's a little bit of memory persistence uh and it did that using artifactory And we'll talk about JFrog a little bit later too, but within a few hours, more than 50 agents started using it to work together.

They hid evidence of cheating, not necessarily from humans, but from the automated scoring processes.

And then I'll go ahead and do the follow up.

Basically there was another thing where, um, kind of like you can, you can think of it as like another CSP bypass where on like Google, you can use like Google scripts to exfiltrate it basically on this same root domain that you actually do have access to like set up an SSRF or not.

Sorry.

I like a request handler.

the best practices in the industry.

Um, and you're right with covering from cross-site scripting, SSRF, um, privilege escalation, a-and et cetera.

But our claim to fame specifically at Novi is the ability to detect, um, novel business logic vulnerabilities.

Uh, I can give you examples.

... uh, when it comes to prompts, but also there is a lot of issues when it comes to...

There's no question that the frontier models are capable of detecting vulnerabilities, but again, when it comes to precision, how many of, of those are really true positive? So the bottleneck really moved from finding issues to being validate what is truly exploitable and what is not.

And I think what is important here, I, I, I think what you're also checking is it's good to have that benchmark, that training gym, just to say, we can find this type of input validation, SSRF, business logic, but you're also testing...

I don't have a good metaphor here for the, for the, a physical gym yet, but what the cost.

It is super easy to run.

Um, it is a form of SSRF, which is a server side request forgery attack.

Yeah.

Um, so for those of you who haven't, like, been familiar with SSRF attacks, an SSRF attack is where I can make a call out to a service, whether it's a web server or in this case a VPN service.

Um, most of the SonicWall VPN services are a web server.

Whatever.

Um, but you go out to a website, and you make a request, and what happens is that server then makes a request to itself on your behalf.

So it looks like it's coming from inside.

which is suggesting attackers are chaining them.

So the unauthenticated SSRF bug for a foothold, and then command injection bug to actually execute code.

And this hits the SMA 1000 series specifically, the 6210, 7210, and 8200 V models.

The separate SMA 100 series and SonicWall's SSL VPN on its firewall are not affected by this, so don't confuse the product lines.

Cover a broad range, not only injection.

Invent realistic situations, weak password reset, unencrypted password storage, outdated library, verbose error messages, IDER on TaskEd, missing logs on failed logins, an SSRF via an import from URL feature, etc.

Step 3.

Map each scenario.