Keeping an eye on AI AI Agents: what are they and what is the problem?

[maxd]

What is an AI agent and why should I care?​

by Max Drayman, Casiinomeister.com

"Anything you can do with a keyboard and a mouse, it can do too."
Hannah Fry,

You do not have permission to view link Log in or register now.

Unless you live under a rock, you'll have heard of AI and chatbots. Even if you don't use Chat GPT or Claude yourself, you'll probably have encountered customer service chatbots. Or you'll have seen the scare headlines to the tune of "AI will kill us all!!!"

Obviously, there's a lot to talk about when it comes to AI. In this article, we're going to focus on agents: a quick overview of what they are, how they're different from your buddy ChatGPT and why you should care.

So, basically, with a chatbot, a human is in control. A human asks the chatbot a question, tells it to do something, checks the results, tells it to do something else. Kind of like playing fetch with a dog: you throw the stick, the dog brings it back, repeat as required. Chatbots, like dogs, sometimes lose the plot and either drop the stick (give you old/incorrect information) or take off after a squirrel (make up something completely).

But an AI agent is something else entirely. You give it a goal and access to some tools and it goes off and figures out how to achieve the goal by itself: it decides which tools to use and how to use them -- on its own. I guess, to extend the dog analogy, it's like if you said to your dog: "I'm bored. Find something nice for me to do" and handed him your phone and your wallet. He books a private jet to Morocco and a weekend in a spa, hands you your travel package and waits patiently for the next command. Oh, and also: the dog bought 400 squeaky toys for himself and all the dogs in the neighbourhood. Just because.

AI Agents : What is happening right now.​

AI Agents are already being used for nefarious reasons and if you haven't been a victim of that yet you certainly will be soon enough. I'm well aware that sounds a bit doomsday-ish but if someone yells at you to get out of the way of a speeding car it's probably wise to move your butt first and ask questions later. At this moment the power that AI Agents have is massively disproportionate to the defences we have against them. Such is the nature of all arms races and we are in the early stages of one now.

A recently reported example in Futurism ("

You do not have permission to view link Log in or register now.

) shows how massive a single operator's attack can be when scaled up in ways that were previously simply out of reach:
"... A cybercriminal used AI agents to commit a string of attacks against more than 100 different companies — an alarming incident that shows how autonomous Agents ... have exponentially raised the threat that a single human hacker can pose."

A recent "round-the-clock cybersecurity emergency" triggered by the South Korean government demonstrates how imbalanced the threat-vs-defences scale really is. The attacker in question breached seven financial institutions "exposing the personal data of more than 65,000 customers" by simply taking an open-source and publicly available penetration-testing system -- ARTEX AI -- "and pointed it at internet-accessible banking portals." The attacker spent 43 hours pillaging the banks before institutional defences were triggered. (Open-Source AI Agent Hacked Seven South Korean Banks).

You do not have permission to view link Log in or register now.

Less dramatic perhaps but equally concerning is how AI agents are being used to apply pressure against public institutions for purely political or commercial motives. EurekaAlert.org recently reported on a case where a UK public consultation on restricting disposable vape sales was overwhelmed with AI-generated submissions targeted at opposing restrictions. Fully 76% of the submissions -- 90,835 of the 118,756 total -- were the result of automated interference. The extent of the interference was revealed "after the bot responses were excluded, 77% of the valid respondents supported restrictions on disposable vape sales—the opposite of what the unfiltered dataset would have indicated."

You do not have permission to view link Log in or register now.

As with (almost) all arms races we will learn, in time, to adapt, re-tool and defend ourselves as needs be. In the meantime we need to be aware of how all this crazy tech can and will affect us.

Conclusion: So why should I care about agents?​

Well, instead of being paralysed by tech industry leaders' warnings of "extinction level events" and "Ai will kill us all" scenarios, we should probably be more focused on how Agentic AI (the cool-kid way to say "AI agents") is affecting our lives right now.

Instead of worrying about the Terminator breaking down your front door and chasing you around the neighbourhood. It's more like he is setting up a heavy metal concert on your front lawn to drown out anything you have to say. Also, he might steal your wallet with your bank cards and ID in it.

Some things we have no control over: bad actors using agents to hack websites and overwhelm services. There we just have to wait for businesses, banks and governments to catch up. But there are still things you can do to reduce the risk of your data being stolen, simply by being more careful about who you give it to.

We've all learned by now not to click on links in emails or text messages, but to go directly to the website and log in. And to use two factor whenever we can.

These days, given the new threats, you should be even more careful about who has your sensitive information. Do you really need to hit "save my payment details" just to save a couple minutes in the checkout process next time? If -- or rather, "when" -- that site gets hacked, they have your name, address and card details. If a shop wants your birthdate to give you a birthday discount, give them your dog's.

Furthermore, don't give your toothbrush, washing machine or toaster your wifi password. Don't tell ChatGPT your secrets or your bank passwords. Don't give your Facebook assistant access to your emails and banking. The less access you allow, the less of a target you are; the less of your sensitive information you give out, the less of a chance it will end up in for sale on the dark web.

In our third and final article in this series we'll be taking a quick look at the mob version of AI Agents: Agent Swarms.

If you missed our first article in the series you'll find that here: AI is turbo-charging iGaming fraud .