A cyberattack campaign utilizing AI agents has compromised over 100 corporate websites, according to Snorre Fagerland of NorForge. The operation, which began on September 25, 2026, involved the use of the open-source platform Cairn to coordinate tasks among AI agents. The attackers launched over 1,100 attack projects, targeting e-commerce sites in the US, Canada, Japan, and Australia.
The agents utilized a software component identified as 'claudecode_deepseek-v4.1-flash', likely combining Anthropic's Claude Code with the DeepSeek V4.1 Flash model. The average duration for a successful breach was just under 23 hours. NorForge estimates the total token expenditure at $5,000, resulting in an average cost of approximately $50 per successful intrusion.
The attackers attempted to breach over 1,200 sites, with an average success rate of about 10%. The goal was to establish persistence within store systems and payment infrastructure. Companies including Deere, SharkNinja, and Cummins were identified as potential targets. While some firms confirmed localized security incidents, no evidence of sensitive data theft has been verified by Reuters.
The Cairn platform, which facilitated the attacks, was found to contain exposed credentials and API keys, with 11% of discovered secrets confirmed as valid by the Trufflehog scanner. The creator of the Cairn platform stated that the software was not intended for malicious use and that such activities violate the platform's original purpose.
