Key Points
- A Chinese-speaking hacker used open-source AI agents to breach more than 100 companies, Norforge said.
- The campaign began September 25 and targeted e-commerce sites in the US, Canada, Australia and Japan.
- The economics matter: autonomous agents put mass intrusion within reach of a single operator.
The latest:
More than 100 companies, including Deere, Sharkninja and Cummins, were breached by a Chinese-speaking hacker running open-source AI agents, Norforge said in an October 19, 2026 report. Snorre Fagerland, the firm’s chief threat researcher, said the agents attempted over 1,200 sites and needed roughly 23 hours to complete a successful intrusion. Norforge is working with Reuters to notify victims.
Details:
- The platform: Norforge said the operator used Cairn, an open-source agent platform, to run intrusion campaigns at scale. Tasks were executed through a worker labeled claudecode_deepseek-v4.1-flash, which Norforge said may indicate Claude Code running on a DeepSeek V4.1 Flash model. The platform’s author, le1xia0, said Cairn was not designed for hacking.
- The economics: Fagerland estimated to Reuters that the full campaign cost around $5,000 in tokens, or roughly $50 per successful breach. The Norforge report put the figure at about $5 per task. Either number places a campaign of this scale within the budget of a single individual rather than a funded group.
- The targets: The campaign started on September 25 and focused on e-commerce sites in the United States, Canada, Australia and Japan, according to Norforge. Agents probed more than 1,200 sites in total, with an average success rate of roughly 10 percent across attempts.
- The infrastructure: Norforge said it identified more than 20 malicious dashboards split into three clusters: a Chinese-language group of 14 dashboards running over 1,100 projects, an English-language group of 10 dashboards aimed at payment cards and cryptocurrency, and a single separate dashboard with 59 projects.
- The loot: The agents harvested hundreds of credentials and keys, according to Norforge, which used the Trufflehog scanning tool to test them. Eleven percent were verified as valid, meaning a substantial portion of the stolen material was live rather than stale or already rotated.
- Deere’s response: Deere said it had taken remediation steps and found no evidence that sensitive information was accessed, according to Norforge and Reuters. The company did not detail which systems were touched or when it first detected the activity.
- The other companies: Sharkninja said the attack hit a standalone support site and was contained. Cummins said it was investigating a possible breach of a site operated by a supplier. Neither company disclosed how many records, if any, were exposed.
- Earlier cases: Gambit Security documented a separate campaign in September that breached 25 companies and stole at least 600,000 payment card records. Taiwan and India have both reported intrusions in which AI agents played a supporting role, pointing to a pattern rather than an isolated incident.
Background:
Cairn is open-source software, meaning anyone can download and run it without vendor approval or oversight. That removes the gatekeeping layer that commercial security and AI tools normally apply, and leaves detection to defenders after the fact.
Between the lines:
The 10 percent success rate looks modest until it is read alongside the 1,200 attempted sites and the $5-per-task cost. Agents that never tire make a low hit rate economically viable, because volume replaces skill. The three separate dashboard clusters, in two languages and aimed at different payoffs, suggest more than one operator has already reached the same conclusion.
What’s next
Watch whether Cummins confirms the supplier-site breach, whether more of the 1,200 probed sites report intrusions as Norforge and Reuters complete victim notifications, and whether Cairn’s maintainers restrict the platform.
