Autonomous Cairn AI agents breach 1,200+ sites via exposed panels, hitting Deere and Cummins supply chains

NorForge researcher Snorre Fagerland located two dozen exposed Cairn control panels by scanning for open instances starting October 2. The AIA0001 cluster alone launched more than 1,100 tasks against retail and manufacturing targets, confirming one hardware vendor compromise through a live web shell and another via memcached session store plus unsafe PHP object handling in legacy Zend Framework. Reuters cross-checked attacker logs to name Deere, SharkNinja, and Cummins among victims, though each reported limited scope and no confirmed data exfiltration.

Cairn's design, intended for authorized testing, allows agents to post findings to a shared board and iterate on short Chinese-language goal strings without continuous oversight. This produced a 10% success rate across varied targets in the US, Japan, and Australia, demonstrating how low-cost token usage can scale persistent access attempts against payment and shop systems. The pattern of exposed tooling mirrors prior open-source pentest frameworks that quickly migrate to criminal use once public.

Official statements from the three named firms emphasize containment on standalone sites or vendor instances, yet independent verification of live shells contradicts claims of negligible impact. No technical attribution ties the operators to state actors; activity remains consistent with profit-driven access brokerage.

Next steps include monitoring for token-cost drops below $20 per breach and similar agent frameworks appearing in procurement logs or dark-web marketplaces within 60 days.