AiStaffo

Hacker Uses AI Agents to Breach More Than 100 Company Websites

Stock photo: office workers reviewing documents

A Chinese-speaking hacker used commercially available AI agents to break into websites of more than 100 companies, including Deere, SharkNinja and Cummins, according to a Reuters investigation published on 19 October 2026. The campaign began on 25 September and mainly targeted e-commerce sites in the United States, Canada, Australia and Japan. The attacker wrote fewer than 100 Chinese characters of instructions and left the agents to do most of the work. An estimated $5,000 in AI usage costs is uncertain, and it is not yet known how much data was taken. For any business that automates work, the lesson is to treat the AI tools and the vendor websites around them as part of the security perimeter.

In short

  • A Chinese-speaking hacker reportedly used off-the-shelf AI agents to target more than 100 company websites, starting 25 September 2026.
  • The cost estimates differ, so treat the $5,000 total as uncertain.
  • Exposed control panels for AI agent platforms were how the campaign was found.
  • Vendor websites and admin panels are the first things to check in any automated workflow.
  • No public evidence yet shows how much data was taken.

What happened

According to a Reuters investigation published on 19 October 2026, a Chinese-speaking hacker used commercially available AI agents to break into websites tied to more than 100 companies. The named victims include farm equipment maker Deere, appliance company SharkNinja and engine maker Cummins. The reported campaign started on 25 September and focused on e-commerce sites in the United States, Canada, Australia and Japan.

The activity came to light through NorForge, a security firm. Its principal threat researcher, Snorre Fagerland, found about two dozen exposed control panels for Cairn, an open-source platform that coordinates AI agents. Fagerland said the hacker gave the agents brief instructions, then let them probe more than 1,200 sites. Successful intrusions typically took about 23 hours of automated testing.

Deere told Reuters that it had taken remediation steps and had no indication that sensitive information was accessed. SharkNinja said the affected site was a standalone product support website and that the threat was contained. Cummins said it was investigating a report about a website run by one of its third-party vendors. Reuters could not establish how much data, if any, was taken. Two smaller businesses also confirmed they had been compromised.

Why it matters for businesses that automate work

The part that should interest operations heads is the cost. Fagerland estimated the campaign used about $5,000 in AI tokens, or roughly $50 per successful intrusion. NorForge later put its own estimate at roughly $5 per task, including failed attempts, and warned that the margin of error is wide. The two figures are not the same measure, and neither is a verified budget. The direction of travel is still clear: one person, a few instructions and a set of agents can test far more doors than a human team could.

The campaign also turned the automation tools themselves into a weak point. The exposed Cairn dashboards reportedly showed targets, findings and, in some cases, stolen credentials. A business that runs its own agents, workflow tools or vendor portals can make the same mistake without meaning to, by leaving a control panel reachable from the open internet.

What changes in practice

Most firms do not need a new security programme this week. They need a short list of the places where automated work touches the public web, and a named person who checks it. The table below sets out four starting points. The steps apply whether you have five staff or five thousand. A smaller firm can often ask its web host or software provider for the answers in a single call. A larger group needs them written into a register.

Exposure pointWhat to checkWho owns it
Admin or control panels for internal toolsConfirm none are reachable from the open internet, and that each requires a loginIT lead or outside provider
Vendor websites that take customer data or ordersAsk the vendor for its latest security test results and a list of who holds admin accessOperations head with procurement
Shop software and pluginsConfirm that security updates are applied on a fixed schedule, not when someone remembersWeb or e-commerce team
AI agent accounts and API keysGive each agent its own key with the narrowest permissions it needs, rotate keys, and keep a log of actionsFinance or IT lead

Two habits matter more than the tools. First, keep a written record of every automated workflow, including the systems it reads and writes. Second, when a vendor is compromised, check what your own automation can still reach through that vendor's login. Many firms learn about this link only after an incident.

Limits

Automation can run the checks, but it cannot decide what risk you accept. A script can list open panels or stale keys. It cannot confirm that a vendor's code is safe, and it cannot tell you whether data has already left. Those judgements still belong to a person who knows the business.

What to watch next

Three things will show how serious the campaign was. The first is whether Cummins confirms the scope of the vendor breach. The second is whether NorForge publishes more on the other exposed instances, which it said were likely malicious. The third is whether the cost estimates converge. Until then, the $5,000 figure should be read as an estimate, not a settled number.

How AiStaffo would automate this

An AiStaffo audit would start by listing every automated workflow and the systems it reads and writes, including vendor portals, shop software and any AI agent dashboards. Our AI workers would then run scheduled checks against that list, flagging any admin page that answers from the open internet, any key without a named owner, and any vendor login that reaches more than one system. The results would go to the operations head as a short report. A person would still decide which fixes to apply and who keeps access. Book a free automation audit.

Questions people ask

Which companies were affected by the AI agent hacking campaign?
According to a Reuters investigation published on 19 October 2026, the named victims include Deere, SharkNinja and Cummins. Reuters identified them by reviewing the hacker's activity, because the researcher's report did not name them. Each company gave a different account of its exposure, so the full list of affected firms is not yet confirmed.
How much did the AI agent attack cost?
The researcher Snorre Fagerland estimated about $5,000 in AI tokens for the whole campaign, or roughly $50 per successful intrusion. NorForge had no actual usage data, so Reuters described the figures as highly uncertain. NorForge's own separate estimate put the cost at roughly $5 per task, including failed attempts.
Was customer data stolen in this campaign?
Not confirmed. Deere said it saw no indication that sensitive information was accessed, and Reuters could not establish how much data, if any, the hackers took. Two smaller businesses did confirm they had been compromised, and a separate hacker observed in the same research reportedly stole credit card data.
What is Cairn and why did it matter?
Cairn is an open-source platform that coordinates AI agents to work through a task. Researchers found it exposed on the internet with control panels that were left open. Those panels showed the attacker's targets and progress, which is how the campaign was traced.
What should a business check after this news?
Start with a list of every internet-facing tool that automated work touches, including admin panels, vendor portals and AI agent accounts. Confirm each one requires a login and that its security updates are current. Ask key vendors for their latest security test results and who holds admin access.
Does this mean automation is not worth using?
No. The campaign shows that automated tools need owners, access limits and regular checks, not that they should be avoided. Automation is less useful where no one knows which systems it touches, so the review should come before any new workflow goes live.

Book a free automation audit

Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.

We use these details only to reply to this request and prepare your audit, and keep them for up to two years. We send no marketing unless you tick the box above. Privacy policy

ai agentscyber securitywebsite securityvendor riske commerce