AI agents as burglars: one hacker breaches 100+ company sites at about $50 each
According to NorForge and Reuters, autonomous AI agents attacked more than 1,200 websites, including sites of Deere, SharkNinja and Cummins. Estimated token bill: $5,000.
Symbolic image · Photo: Christopher Bowns / Wikimedia Commons, CC BY-SA 2.0 (cropped and resized)
A suspected Chinese-speaking attacker has used autonomous AI agents to target more than 1,200 websites since September 25, compromising over 100 of them. That is the finding of a report published on Monday by researcher Snorre Fagerland of security firm NorForge, which Reuters reviewed in advance. Targets were mainly online shops in the US, Canada, Japan and Australia, with the apparent goal of lasting access to shop and payment systems.
Fagerland found exposed control panels of the open-source agent platform Cairn. One worker was named “claudecode_deepseek-v4.1-flash”, which NorForge suggests could point to Anthropic's Claude Code running DeepSeek's V4.1 Flash model; this has not been confirmed. Successful intrusions took just under 23 hours on average; Fagerland roughly estimates token costs at $5,000, or about $50 per successful hack. “It's not script kiddie stuff,” he told Reuters.
Deere said it had taken remediation steps and saw no sign that sensitive information was accessed. SharkNinja said a standalone product support site was hit, while Cummins is looking into a possible attack on a vendor-run website. Reuters could not establish whether any data was taken. Cairn's author said on X the platform was not designed for hacking attacks.