Remember when "script kiddie" was an insult? Meet its 2026 upgrade: a single operator who hands a fleet of AI agents a target list, goes to bed, and wakes up with web shells. The economics of hacking just got a very uncomfortable discount.

1,200 Targets, 23 Hours, One Shared Whiteboard

Security firm NorForge says a Chinese-speaking hacker used AI agents to attack more than 1,200 websites across the US, Canada, Japan and Australia, breaking into more than 100 of them. The agents ran on Cairn, an open-source multi-agent framework where agents post findings to a shared board and grind toward a goal step by step; the campaign began September 25, and successful attacks took just under 23 hours on average. Reuters independently identified victims including sites tied to Deere, SharkNinja and Cummins, though none of the three confirmed that sensitive data was taken.

NorForge's principal threat researcher estimated roughly $5,000 in AI token spend, or about $50 per successful hack, a figure he called rough. The agents mostly hunted online shops, trying to gain code execution and plant a web shell, apparently for lasting access to shop and payment systems. In one case, finding nothing wrong with a big company's public app, the agents pivoted to a linked control panel and, after hours of trial and error, got its server to run commands.

The Attackers Left the Lights On

Here's the delicious irony: NorForge only spotted this because operators left their Cairn control panels exposed on the internet. Those panels leaked hundreds of credentials and API keys, 11% of which checked out as valid. Even AI-powered criminals, it turns out, forget to lock the admin screen.

The bigger takeaway is persistence, not genius. These agents don't get bored, don't take lunch, and will cheerfully hammer your forgotten support site for a full day. Notice that SharkNinja's incident hit a standalone product support site and Cummins is looking into a site run by one of its vendors: the attack surface that bites you is usually the one nobody remembers owning. Cairn's author has said the tool wasn't designed for hacking, which is true of most good tools right up until it isn't.

When an intrusion costs less than a nice dinner, "we're too small to be a target" is officially retired.

If the idea of an AI swarm patiently probing your forgotten subdomains makes you twitch, our team can map your real attack surface and harden it before someone else's agents do the mapping for you, so start the conversation here.

Source: The Next Web