Unprotected AI Models Targeted by Hackers for Computing Power Theft
Security researchers from SentinelOne and Censys have uncovered a critical vulnerability in private AI servers, with approximately 175,000 systems exposed to the public internet. These servers, running open-source software Ollama, are being exploited by hackers to hijack computing resources for illicit activities.
The study tracked 7.23 million observations over 300 days, revealing that 23,000 systems remain persistently online, making them prime targets for unauthorized use. Notably, 18% of exposed US systems are concentrated in Virginia's data center hub, while Beijing hosts 30% of China's vulnerable infrastructure.
This security lapse highlights growing concerns about decentralized AI infrastructure management, particularly as organizations increasingly adopt open-source models like Meta's Llama and Google's Gemma for private deployments.
