Hackers Deploy Autonomous AI Agents to Breach Corporate Websites
Criminals leverage open-source AI tools to automate vulnerability scanning and system exploitation.
1Autonomous AI Attack Campaign
A Chinese-speaking criminal actor utilized autonomous AI agents to target over 1,200 websites. The campaign, identified by researcher Snorre Fagerland of NorForge, exploited exposed instances of the open-source tool Cairn. These agents operated with minimal human intervention, successfully compromising systems belonging to major companies like Deere, SharkNinja, and Cummins.
2Cairn Tool Misuse Explained
Cairn is an open-source AI framework originally designed for authorized penetration testing. It allows operators to set goals for agents, which then execute tasks step-by-step. In this malicious campaign, operators pointed these agents at online shops without permission, using them to identify vulnerabilities and plant web shells for persistent access.
3Operational Scale and Efficiency
NorForge identified three clusters of exposed control panels, with the busiest, AIA0001, linked to over 1,100 attack projects. Successful breaches took an average of 23 hours to complete. The cost of these operations was remarkably low, estimated at approximately $50 in token usage per successful hack across the campaign.
4Technical Execution and Vulnerabilities
The agents employed varied methods depending on the target, including exploiting memcached session stores and outdated Zend Framework code. In one instance, agents identified a single cookie setting to bypass security on a large US retailer's backend. These diverse attack paths demonstrate the agents' ability to adapt to specific site architectures.
5Exposed Control Panel Risks
The security breach was discovered because attackers left their Cairn control panels accessible on the public internet. These panels exposed real-time logs, stolen credentials, and full project controls. NorForge used the Trufflehog scanner to find hundreds of secrets within these panels, verifying that 11% of the exposed credentials remained valid.
6Victim Impact and Response
Affected companies reported limited damage following the disclosures. Deere confirmed it implemented fixes with no evidence of sensitive data loss, while SharkNinja contained the threat on a standalone support site. Cummins is currently investigating reports regarding a vendor-managed website. Two smaller businesses also confirmed they were compromised during the campaign.
7Broader Security Industry Implications
The exposure of penetration testing tools is not limited to criminal actors. NorForge discovered legitimate security firms also leaving client test data on unsecured panels. This highlights a systemic failure in managing AI-driven security infrastructure, where the tools intended to protect systems become significant liabilities when left exposed to the public.
8Future of AI-Driven Threats
This campaign demonstrates the shift toward low-cost, automated exploitation using AI models like DeepSeek and Claude Code. As agents become more capable of autonomous decision-making, organizations must prioritize attack surface visibility and strict permission controls. Securing AI infrastructure is now as critical as defending the primary production environments they target.
Original article: The Hacker News — Hackers Deploy Autonomous AI Agents to Breach Corporate Websites