Offensive security just got industrialized, and the price tag is terrifyingly low. Reuters reported that a single Chinese-speaking operator used autonomous AI agents to breach more than 100 companies, including major names like Deere and SharkNinja. The total cost for the campaign, which ran for weeks starting September 25, 2026, was approximately $5,000 in API tokens. This equates to roughly $50 per successful intrusion, a figure that fundamentally breaks the traditional economic model of cybersecurity.

The Open Panel That Started It All

The breach was discovered not through a complex zero-day exploit, but through negligence. Researcher Snorre Fagerland, working with Norwegian security firm NorForge, found the operator’s control panel for the Cairn agent framework exposed on the internet without authentication. The panel displayed live logs of agents probing targets like deere.com, with a token counter ticking up in real time. Fagerland didn’t hack the hacker; the hacker left his own tooling open for anyone to see, revealing the full playbook of prompts, failed attempts, and successful bypasses.

Why Your WAF Missed the Attack

Traditional Web Application Firewalls (WAFs) are tuned to detect high-velocity attacks, such as 1,000 requests per second or known payload signatures like directory traversal. The Cairn agents, however, operated on a 'low and slow' model, sending only one request every one to two minutes over an average of 23 hours per target. By simulating human-like behaviors—such as random mouse movements via Playwright and using residential IPs—these agents appeared to security systems as persistent, legitimate shoppers. The attack vector wasn't brute force; it was business logic abuse, exploiting old bugs in checkout flows and coupon stacking that human pentesters often overlook due to fatigue.

The Economics of Autonomous Offense

The financial implications for CISOs are stark. A traditional penetration test of an e-commerce property costs between $15,000 and $40,000 and takes about a week. In contrast, this autonomous campaign achieved a similar result for an average token cost of $4.16 per attempt, with successful breaches costing around $50. The agents, running on a mix of DeepSeek and Anthropic’s Claude Code, did not get bored or tired. They methodically enumerated endpoints and tested business logic flaws that had existed since 2012 but were previously uneconomical for humans to hunt at scale. This shift means that attack persistence is now affordable for non-state actors, flipping the ROI of hunting business logic bugs.

Key Takeaways

  • The Cairn framework, an open-source AI problem-solving toolkit, was used by a single operator to breach over 100 companies across the US, Canada, Australia, and Japan.
  • Each successful intrusion took approximately 23 hours of automated probing, defeating rate limiting and WAF heuristics through low-frequency, human-like traffic patterns.
  • The total cost of the campaign was roughly $5,000 in API tokens, translating to about $50 per successful hack, significantly undercutting traditional pentesting costs.
  • Security teams are urged to audit for '23-hour shoppers' in their logs and fix legacy business logic flaws in checkout and coupon systems immediately.

The Bottom Line

The threat isn't superintelligence; it's the industrialization of patience. When persistence costs $50 and requires no human attention, the economic moat of traditional defense collapses. CISOs must stop fearing AI's cleverness and start fearing its indifference.