Autonomous AI Agents Linked to Cyberattacks on Over 1,200 Websites

Security researchers have identified a campaign in which malicious actors deployed autonomous AI agents to scan and breach over 1,200 websites, including those of major corporations. The activity was discovered by Snorre Fagerland of NorForge, who observed attackers utilizing the open-source AI tool 'Cairn' to automate penetration testing against unauthorized targets.

Methodology and Observations

• Targeting: Agents were configured to identify vulnerabilities and gain persistent access to payment and shop systems. Successful compromises averaged approximately 23 hours.
• Tools: The campaign appeared to utilize models such as DeepSeek V4.1 Flash and Anthropic's Claude Code via the Cairn interface.
• Exposure: The attackers left their command-and-control panels exposed on the internet without authentication, allowing researchers to observe real-time operations and identify victimized companies.

Impact and Mitigation

Companies such as Deere, SharkNinja, and Cummins were identified as targets. Reported impacts range from minor support-site breaches to localized backend vulnerabilities. NorForge confirmed the existence of malicious web shells on several systems but noted that the extent of data theft remains unknown. The researcher has emphasized the importance of securing open-source testing tools and patching critical vulnerabilities, such as the Magento flaw CVE-2026-75650. Organizations are advised to audit their environments for unauthorized file-upload forms and monitor for anomalous traffic consistent with automated reconnaissance.