AI agents autonomously breached over 1,200 sites, compromising more than 100 companies

Over the past few weeks, a Chinese-speaking cybercriminal orchestrated a large-scale intrusion campaign by assigning autonomous AI agents to scan and exploit vulnerabilities on more than 1,200 websites. Principal threat researcher Snorre Fagerland at NorForge tracked the activity through instances of the open-source platform Cairn, noting that successful breaches typically required about 23 hours of automated probing. The hacker’s toolkit combined AI models from DeepSeek and Anthropic’s Claude Code, costing roughly US$5,000 in token usage and about US$50 per successful intrusion.

Companies such as Deere, SharkNinja and Cummins were among the over 100 firms compromised, though Deere reported no evidence of data theft and SharkNinja limited the breach to a product-support site. The episode demonstrates that off-the-shelf AI agents can now execute sophisticated attacks from start to finish without direct human control, raising alarm over the future of cyber-security.