One Hacker Breached 100+ Sites With AI Agents for ~$5,000: The Attack Economy Changed | inicIA
Direct answer
On October 19, 2026, Reuters published an exclusive on the investigation by researcher Snorre Fagerland of cybersecurity firm NorForge: a Chinese-speaking hacker used AI agents to breach the websites of more than 100 companies over recent weeks. The agents attempted intrusions against more than 1,200 sites in the United States, Canada, Australia and Japan; victims identified by Reuters include Deere (farm equipment), SharkNinja (home appliances) and Cummins (engines). Fagerland discovered the campaign by finding about two dozen exposed installations of Cairn, an open-source AI platform popular with hackers, and was able to observe the attacks as they happened starting in early October. The numbers that matter: successful intrusions took on average about 23 hours of autonomous agent work, and the entire campaign cost an estimated ~$5,000 in AI tokens —about $50 per successful breach. The agents reportedly used a mix of DeepSeek and Anthropic's Claude Code, though researchers could not confirm the setup. Reuters could not establish whether data was exfiltrated. Deere said it addressed the incident with no evidence of sensitive information access; SharkNinja said the breach was limited to a separate support site and was contained; Cummins said it is investigating a possible attack on a third-party vendor's website. (Sources: Reuters, The Latent, CyberSecAsia.)
Why this case is different
- The hacker barely worked: the attacker's instructions were under 100 Chinese characters; the agents investigated targets, searched for vulnerabilities and attempted entry autonomously for hours. In one case, the agents found nothing in the main application, explored connected infrastructure and discovered an exposed control panel with a subtle data-processing vulnerability.
- The cost is the headline: $50 per successful breach is less than one hour of commercial pentesting consulting. Complex attacks stopped being expensive.
- "It's a new world now": Fagerland's quote summarizes the real-world evidence that commercially available agents can pull off complex intrusions end to end with minimal human intervention.
- The investigation's honesty: no confirmed attacker identity, no confirmed evidence of data theft, and the cost is a rough estimate. Serious coverage says so; sensational coverage does not.
What your company should do this week
- Automate your defense to the attack's level: if attacking costs $50 per site, your manual monitoring already lost. Continuous detection tools and defensive AI stop being a luxury.
- Review your exposed surface: the campaign was discovered through control panels exposed on the internet. Audit what of your company is publicly visible without protection.
- Outside vendors count: Cummins was reached via a third party's site. Your supply chain is your perimeter.
- Patching speed: if your remediation cycle is measured in weeks and the attacker works in hours, the gap is your real risk.
- Don't buy panic, buy process: the case does not prove "AI breaks everything"; it proves the attack industrialized. The answer is industrializing defense: monitoring, containment and response with the same automation.
Related links
- Microsoft: your AI agents are the new insider threat — Guardian Agents and live permissions
- Bank of America hired 1,000 people to fix what AI found: the hidden cost
- The CEO deepfake on a video call that cost $25.6M: the protocol that would have prevented it
Frequently asked questions
What is Cairn?
An open-source AI platform for solving problems with agents: you give it a goal and the agents work step by step. It is legitimate technology, but hackers also use it to automate attacks.
Was company data stolen?
Reuters could not establish whether the breaches resulted in unauthorized data extraction. Deere said it had no evidence of sensitive information access; SharkNinja contained the incident to a separate support site.
Which AI models did the attacker use?
NorForge estimated it may have been a mix of DeepSeek and Anthropic's Claude Code, but researchers could not confirm the exact setup.
Is my small business at risk?
The campaign targeted more than 1,200 sites with indiscriminate vulnerability scanning, not hand-picked targets. Any site with an exposed weakness can be found by an automated sweep.
Take the next step
AI news only matters if you know what to do with it in your business. inicIA helps you turn every market move into concrete decisions for your company. WhatsApp: https://wa.me/17866120930
Chat on WhatsAppBy inicIA
