Single Hacker Breaches 100+ Sites Using AI Agents, NorForge Report Finds
A lone attacker used AI tools to compromise over 100 e-commerce sites, costing roughly $5,000 in token fees, according to a new analysis by NorForge.
Key points
- More than 100 websites were successfully breached by a single hacker using AI agents.
- The campaign targeted e-commerce sites in the US, Canada, Australia, and Japan starting September 25.
- NorForge estimates the operation cost approximately $5,000 in AI token usage.
A lone hacker successfully compromised more than 100 websites belonging to businesses of various sizes over the last few weeks, stealing sensitive data including credit card numbers and cryptocurrency. According to Reuters, citing a recent report from cybersecurity firm NorForge, the attacker relied on AI agents rather than a traditional hacking collective to execute the breaches.
The NorForge report, published on Monday, details how an unnamed Chinese-speaking hacker outsourced nearly all breach attempts to AI agents. These agents utilized Cairn, an open-source AI problem-solving engine available on GitHub, to identify and exploit weaknesses in systems with lax cybersecurity. The platform allows users to coordinate agents for specific tasks, such as finding vulnerabilities or breaking into systems.
Operational Details and Costs
Security researchers observed one exposed operation involving this hacker launching a campaign against e-commerce sites in the United States, Canada, Australia, and Japan. The spree began on September 25 and attempted to hit more than 1,200 sites by last week. While only about 10 percent of these attempts resulted in successful breaches, that failure rate still equates to a significant number of compromised systems.
The campaign required minimal technical training or tuning. Per Reuters, the hacker issued instructions in fewer than 100 Chinese characters. NorForge noted that some prompts were as simple as “fetch credit cards” or “fetch crypto.” The AI agents completed these tasks rapidly, with breaches occurring in under a day and an average completion time of about 23 hours.
Underlying Technology and Exposure
Cairn served as the harness for these operations, but the underlying AI models were largely a mix of China’s DeepSeek and Anthropic’s Claude Code. NorForge estimated that the widespread hacking effort that poked holes in over 1,000 sites cost about $5,000 in tokens to operate. This low cost highlights the efficiency of the automated approach compared to traditional manual hacking methods.
Many hackers have left their Cairn panels exposed, allowing security researchers to monitor their actions. NorForge was able to identify these exposed operations and observe the prompt-and-response interactions between the hackers and the AI models. This visibility provided insight into the non-technical nature of the commands used to drive the attacks.
What should you do?
Security teams should monitor for unusual AI-driven activity and ensure their systems are not exposing management panels to the public. Organizations should review their e-commerce platforms for known vulnerabilities that could be exploited by automated agents. Vigilance is key, as the low barrier to entry for such attacks means more actors may attempt similar campaigns.
What to do and how to stay safe: NorForge
- Audit exposed administrative panels and APIs to ensure they are not accessible to unauthorized users or public internet scans.
- Implement strict rate limiting and behavioral analysis to detect automated scanning patterns typical of AI-driven reconnaissance.
- Review access logs for anomalies that suggest automated credential stuffing or data exfiltration attempts from unknown sources.
- Ensure all e-commerce components are patched and configured securely to resist common vulnerability exploitation techniques.
Step-by-step guide: Mobile Device Security: How to Protect Data Beyond Passwords
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How much did the hacker spend to conduct this campaign?
NorForge estimated the widespread hacking effort that poked holes in over 1,000 sites cost about $5,000 in tokens.
Which AI models were used in the attacks?
The underlying models were largely a mix of China’s DeepSeek and Anthropic’s Claude Code.
How many sites were successfully breached?
More than 100 websites were successfully hit, though the campaign attempted to breach over 1,200 sites.
