New technology is always hailed as something that will elevate humanity’s finest instincts. When the phonograph was invented, it was thought we’d all sit around listening to Beethoven. Ditto with the telephone, radio, television and so forth.
Even things as boring and soon forgotten as the 976 prefix (which let scammers and the occasional legit business charge stuff to your phone number) were mostly used for nefarious ends before eventually going mainstream. Oh, we forgot to mention porn, which lives on in just about every medium imaginable.
And so it is with AI. Artificial intelligence will let us all retire (i.e., put us out of work) and elevate humanity to a higher plane. And AI agents will do so with even less input from their human masters.
Bur first, before that all happens, they’ll perform a little skulduggery, just to get the kinks out and iron out any initial problems with the technology.
An agent like no other
When you hear the word “agent,” you may think of a human investigator from the FBI, Secret Service or the cursed ICE. But our topic today is the AI agent, a big chunk of software that will take on AI tasks and execute them, all on its own, while its human employers loll about contemplating the meaning of life and how to better chisel the tax collector.
Until the other day, cybercriminals were using artificial intelligence to write phishing emails, generate malicious code and automate routine hacking tasks but now, a human trickster has unleashed a band of autonomous AI agents who have hacked into at least 1,200 websites, successfully breaching at least 100 of them. And they’re just getting started.
A Reuters investigation revealed all this misbegotten misuse of AI. It identified victims including Deere, SharkNinja and Cummins, none of whom had much to say about the matter.
What’s significant about this is that it represents a shift from AI-assisted hacking to AI-operated hacking. It’s basically what happens when you train a young assistant to break into safes and a few days later, he takes off on his own, emptying safes all over town.
A very efficient operation
One of AI’s promises is that it is incredibly efficient and can accomplish an amazing amount of work with very little human input (never mind the electricity and water required).
And so it is with AI-operated hacking. It’s very efficient. In this case, the instructions amounted to less than 100 Chinese characters, according to Reuters. From there, the agents took off on their own, attacking systems in the United States, Canada, Australia and Japan.
AI, as you may know, operates on tokens. They’re sort of like subway tokens — a single token will get you x actions by your AI system of choice. In this case, the AI crooks used about $5,000 worth of tokens, not bad for a herculean task that would eat up hundreds or thousands of human hacker hours (h3).
We could go on about this but the salient point is that what we have here is not only AI-controlled hacking but also very economical hacking. These agents could probably hack everything on earth without going over the limit on their credit cards.
There’s a lot more to be said about this, but much of it is rather technical and not really all that interesting so we assigned it to a companion site, ConsumerNews.ai, which uses — what else? — AI to keep an eye on the AI world. You can read its report here.
In parting, let us remind you that the legwork, to use a hopelessly outmoded term, was carried out by Reuters