Study finds more than 22,000 uncensored AI models on hosting platforms
More than 22,000 uncensored large language models, models altered so they are less likely to refuse harmful requests, are freely available on major model hosting platforms, and few carry any warning, according to a study by researchers at the University of Illinois Urbana-Champaign and Nanyang Technological University.
The researchers analysed 22,096 such models collected from Hugging Face, Kaggle and John Snow Labs in the US and OpenCSG and ModelScope in China, identifying more than 11,000 that appeared across multiple platforms, with Hugging Face the main upstream source. Only about 13% carried moderation warnings, and security threat indicators were not consistently kept as models were copied between platforms. "Models can travel more easily than the information needed to assess their risks," co-author Zilong Lin said.
Twenty-seven of the models, downloaded 136,000 times, showed threat indicators in security scans, including malware signatures and unsafe code that could allow arbitrary code execution. The number of new uncensored models has grown from about 100 a month in May 2023 to more than 1,000 a month, and one, dolphin-2.9.1-yi-1.5-34b, has more than 53 million downloads. The models tested were generally more likely than standard models to comply with harmful requests across 11 categories, including child abuse, illegal activity and fraud, according to The Register.
We hope you enjoyed this article
Consider subscribing to one of our newsletters like Cybersecurity AI Weekly or Daily AI Brief.
Also, consider following us on social media:
More from Cybersecurity
Nov 19 Beazley Security finds ransomware up 38% and an attack run almost entirely by AI agents Nov 19 Google Labs introduces Stitch Loop, an agent that keeps improving shipped products Nov 19 UK regulators set out how financial firms should handle AI-driven cyber risk Nov 19 Arizona opens criminal investigation into OpenAI over what it calls rogue AI agents in banks Nov 19 Citi tells customers they are on their own if third-party AI agents misuse their accountsCybersecurity AI Weekly
Weekly newsletter about AI in Cybersecurity.
Humanoid Robot Makers
Companies building humanoid robots, from startups to Tesla and Boston Dynamics
Industry analysis
2025 Global Business Services Agenda: Gen AI Takes Center Stage
This industry analysis by The Hackett Group explores the transformative impact of generative artificial intelligence (Gen AI) on global business services (GBS) in 2025. The study highlights the shift from exploration to acceleration of Gen AI initiatives, with 89% of executives advancing these projects to improve customer satisfaction, innovate products, and reduce costs. The report also discusses the challenges and strategies for successful Gen AI adoption, emphasizing the need for a technology-enabled operating model and the importance of reskilling the workforce.
Read moreYou may also like
Pluto Security finds cryptominers in Hugging Face model repositories
ENISA: frontier AI models found 220 potential security flaws in EU institutions' open-source code
OpenAI details how it contains frontier models, including a pause after a DNS exploit
Microsoft research shows backdoored models can hijack coding agents
Hugging Face Malware Campaign Returns After Takedown
Daily AI Brief: the AI news that matters, in your inbox.