Researcher says one hacker used AI agents to break into more than 100 company websites for about $5,000
NorForge found the campaign through control panels left open online of an open-source agent tool. Deere, SharkNinja and Cummins responded to Reuters; how much data was taken is not known.
Illustration · not a photoHow this story was checked
- ResearchedWith AI assistance · 2 sources
- AI source checkEach sentence matched to its source
- Editor approvalMayank Sahu · 20 Oct, 12:01
- Published20 Oct, 12:01
Verify this story
- 2 sources, 0 official · 2 with an archived copy · see sources
- Every fact has a label and a source number · fact ledger
- What we searched, read and could not verify · research note
- Machine-readable: facts as JSON · text as Markdown
How this was made: researched, drafted and source-checked by AI (Claude, made by Anthropic), then published under the editor’s standing approval. A person did not read it line by line before it went live. AI policy
In short
A single Chinese-speaking hacker used AI agents to break into the websites of more than 100 companies, spending about $5,000 on AI tokens, according to NorForge researcher Snorre Fagerland, whose findings Reuters reported on 19 October 2026. The agents targeted more than 1,200 sites.
What the researcher found
Snorre Fagerland, principal threat researcher at the security firm NorForge, found the campaign by scanning the web for control panels accidentally left open online of Cairn, an open-source platform that runs teams of AI agents, Reuters reported.12 From about 2 October he found roughly two dozen panels being used for malicious work.1
The main operator, who wrote in Chinese, broke into more than 100 company websites, and the agents targeted more than 1,200 sites, according to Reuters.12 The campaign began on 25 September, aimed at e-commerce sites in the US, Canada, Australia and Japan.1 The operator’s instructions were short, under 100 Chinese characters, and the agents did most of the work.1
The cost
Fagerland estimates the hacker spent about $5,000 on AI tokens, roughly $50 per successful break-in.12 NorForge’s report puts the average success rate at about 10% and says successful attacks took just under 23 hours.2 These are the researcher’s estimates.
Which tools
Fagerland assessed that the campaign “appeared to be using” a mix of DeepSeek models and Anthropic’s Claude Code; TNW reports a worker labelled “claudecode_deepseek-v4.1-flash”, which NorForge says could mean Claude Code running a DeepSeek model.12 Cairn’s author said the platform was not designed for hacking.12
What the companies said
Reuters identified some victims itself.1 Deere said it took “necessary remediation actions” and saw no sign sensitive information was accessed. SharkNinja said a standalone product-support site was hit and the threat was contained. Cummins said it was investigating a possible attack on a site run by a vendor.12 Reuters could not establish how much data, if any, was taken.1
Why it matters
Security researchers have warned that agents lower the cost of attacks. This case gives one of the first detailed price tags. Basic defences, such as patching and keeping admin panels off the public internet, still matter most. See also our report on CERT-In’s AI agent advisory.
Disclosure: Anawords uses Claude, made by Anthropic, as one of its research tools. Anthropic has no say in our coverage.
Fact ledger
How labels workTracked from this story
All trackersMultiple companies (hacking campaign), Cairn open-source agent platform, misused · started not known · duration not published
All incidentsQuestions readers ask
How did the hacker use AI agents?
According to NorForge, the hacker gave short instructions to an open-source multi-agent tool, Cairn, and the agents carried out most of the intrusion steps.
Which companies were affected?
Reuters named Deere, SharkNinja and Cummins, which each responded. Other victims were not named or did not respond.
Was customer data stolen?
Reuters could not establish how much data, if any, was taken. Deere said it saw no sign sensitive information was accessed.