I used to love collecting and making reference books. Some of them sat next to me, or on my closest bookshelf, for years.

That's no longer the case. If I need to troubleshoot something or understand how it works, I turn to the multitude of AI models to help me solve it.

As humans we don't really learn until something catastrophic happens to us, or to an organisation we're part of. We're learners through experience.

But sometimes a story comes along that's compelling enough to catch our attention anyway. Here's one.

The Hacker News reported on a "hacker" who spun up a rented server, set up an unprompted Hermes AI agent, and pointed it at Thailand's Ministry of Finance for post-exploitation.

Let's draw a distinction here.

The attacker set this up for post-exploitation, which means they already had access to the network. Based on the article, the Hermes agent was sweeping for elevated-permission binaries, which suggests they didn't start with an administrator-level account.

So this was a valid account with normal user permissions. Even from that starting point, it found a folder that "held Office documents, performance evaluations, and personnel records dating to 2012. The logs show the agent reading the directory. None of them show the files leaving it."

If it can read it, it could likely have transmitted it too. The article just specifies that nothing was transferred, so it's hard to infer intent either way. I'd consider the data leaked the moment the files were read.

More on this here: https://hunt.io/blog/thailand-ministry-finance-targeted-with-hermes-ai-agent​

What I want to highlight is this: if a valid user can use Hermes to run an attack from a remote server, imagine what an insider could do by just installing Hermes on their corporate laptop and letting it run.

You might think that's an impossible scenario. Why would someone risk that kind of exposure?

But maybe they're already on their way out. Maybe they're about to be fired, or made redundant. Or maybe they're a remote worker you've hired in a different country, one with no extradition agreement with yours.

These are the insider threat risks organisations now face when it comes to AI.

I spend a lot of time digging into these. I have a workshop coming up where I show exactly how easy it is for the average person to run this maliciously inside a corporate environment.

Come be part of my vetted insider risk community. Our next session is next week: https://www.dtex.ai/events/ita-workshop-hermes-ai-agent-detection/​