The store where you shop, your company’s website, or a service holding your information does not have to be a high-profile target to face a cyberattack. When one person can assign several AI assistants to search for weaknesses for hours, trying to break into large numbers of websites becomes more affordable.
A campaign described by Reuters and covered by Xataka illustrates that shift: more than 1,200 websites faced intrusion attempts, and over 100 were compromised. That does not mean information was stolen from every site. The extent of the access and its consequences remain unclear.
A short instruction, a lot of delegated work
According to the reporting, the attacker used instructions of fewer than 100 Chinese characters to set the agents to work. That is a striking detail, but this was not a magic phrase capable of unlocking any server. Behind it was a system that could divide up tasks, use tools, and build on the results of earlier attempts.
The platform, Cairn, is an open problem-solving system involving multiple AI agents, rather than a tool designed specifically for criminal activity. In this campaign, its capabilities were used to search for weaknesses and explore possible entry points without requiring fresh human directions at every step.
NorForge researcher Snorre Fagerland documented one case in which the agents failed to find an entry point through a company’s main application and expanded their search to related infrastructure. After hours of testing, they reported that they had managed to execute commands on a server.
The economics matter as much as the technology
Fagerland estimated that the campaign consumed about $5,000 in AI model usage, or roughly $50 per compromised website. Those are estimates, not amounts established through invoices, and they do not necessarily represent the operation’s full cost.
According to NorForge’s analysis, successful projects took just under 23 hours on average. The important distinction is that those hours did not require a person’s continuous attention. An attacker could keep multiple attempts running while the agents worked independently.
What this means for your information and your business
The risk is not that AI has made every defense useless. It is that attackers can search for and exploit security lapses on a larger scale. For a small business, an outdated website or a poorly protected administration panel may become an easier opportunity to discover.
Basic precautions still matter:
- Keep internet-facing applications and services up to date.
- Enable two-factor authentication and avoid reusing passwords.
- Restrict permissions for accounts and AI assistants connected to internal systems.
- Maintain backups and review suspicious-activity alerts.
These are general preventive measures, not a confirmed explanation of how every intrusion in this campaign occurred.
What remains unknown
Reuters identified websites linked to Deere, SharkNinja, and Cummins among those affected, but could not establish how much information, if any, was taken without authorization. Some companies described a limited impact. The language used by the attacker should not be treated as proof of nationality or government involvement, either.
The takeaway is not that cybercrime no longer needs people. It is that one person can delegate more of the work. For the teams protecting the services we use every day, that means facing more attempts without counting on an attacker running out of time.
