Hello and welcome to the Tuesday, July 14th, 2026 edition of the Sands in the Storm Center's Stormcast.

My name is Johannes Ulrich, recorded a day from Washington, DC.

And this episode is brought you by the Sands.edu graduate certificate program in cyber security

engineering. Manuel today wrote up an interesting trend that he has been observing in a site that

he is watching. Well, it's just a basic web server, but more and more, it is being hit by requests for MCP connections, so the

model control protocol that's commonly used with AI, as well as requests for AI credentials.

Now for a long time we have seen a lot of requests for like no dot ENV files and the like

essentially files that typically contain credentials.

But what Manuel is seeing and that sort of matches our global data as well, more and more

of these scans are looking for credentials specifically associated with AI models.

With the goal, of course, being that some of these credentials could be abused by the attacker to gain access to free to the attacker AI tokens.

Now, the MCP connection attempts are also interesting.

First of all, many of them are just checking for the presence of the slash MCP URL.

Of course, if there's a 404 coming back, the URL doesn't exist,

and the attacker knows that this is not at least the right access for the API.

Well, but a couple of requests that Manuel observed also contained the full MCP-compliant

payload to initiate new connections. So definitely attackers are looking for these services to again likely abuse them for their own purposes.

And SISA, together with a few other self-information security related federal organizations and the the FBI has published some guidance on

how to better protect your routers.

Now, the routers we're talking about here are more sort of the ISP enterprise routers, not

your home router, which of course is often the target of attacks.

The guidance here is specific to exploits that have been observed

...