A security researcher has watched attackers deploy autonomous AI agents against more than 1,200 retail and service websites, with at least 100 organisations affected. The activity centred on Cairn, an open‑source AI tool, whose exposed control panels were found and scanned by NorForge. The operation appears to have begun in early October, with around two dozen Cairn instances seen in use, and a cluster labelled AIA0001 driving the most traffic—linking 14 panels to more than 1,100 attack projects.
The campaign’s stated aim was persistent access to shop and payment systems, targeting victims in the United States, Japan and Australia. Victims named by Reuters include Deere & Company, SharkNinja and Cummins, though the defenders say the damage was limited and there is no confirmed disclosure of sensitive data.
NorForge’s investigation details how the agents operated: a typical task instructed the AI to find a way to run code on a shop and then plant a web shell uploader, a foothold that could be swapped for more dangerous capabilities. One documented intrusion involved unauthorized write access to a memcached session store, insecure handling of saved PHP objects, and an old Zend Framework weakness. In one large US retailer, a single cookie setting exposed a vulnerable backend server.
The operation ran at low cost—NorForge estimates roughly $5 per task, with total campaign expenditure around $5,000 in tokens (about $50 per successful hack). The panels themselves expose sensitive data such as credentials and keys, and a separate English-language cluster leaked operator controls. Evidence points to the campaign using DeepSeek and Claude Code models, though Reuters has not confirmed specifics. Defenders are urged to kill exposed panels and review Magento patches (CVE-2026-75650) for affected builds.