{
  "id": 20734623,
  "title": "Uncensored AI models are thriving on code hosting platforms",
  "url": "https://urgent.news/2026/11/19/uncensored-ai-models-are-thriving-on-code-hosting-platforms",
  "topic": "ai",
  "section": "AI",
  "published": "2026-11-19T09:30:00.000Z",
  "source": {
    "name": "The Register Science",
    "slug": "the-register-science",
    "url": "https://www.theregister.com/ai-and-ml/2026/11/19/uncensored-ai-models-are-thriving-on-code-hosting-platforms/5310891"
  },
  "original_language": "en",
  "account": "Uncensored large language models (ULLMs), which can generate harmful content, are easily accessible on popular model-hosting websites, raising concerns about the commitment of the AI industry to safety. Researchers from the University of Illinois Urbana-Champaign and Nanyang Technological University in Singapore analyzed 22,096 ULLMs from five major LLM hosting platforms. They found that uncensored LLMs are readily available on platforms such as Hugging Face, Kaggle, OpenCSG, and ModelScope. These unrestrained models have been integrated into applications, GitHub projects, and model routing platforms, making it difficult for users to know the nature of the underlying LLM. Uncensored LLMs lack content filtering or refusal safeguards, allowing them to respond to requests that more aligned models might refuse. This poses a challenge for assessing the safety of LLMs, as evaluating their safety typically requires deploying the model and testing its responses to potentially harmful prompts, which can be computationally expensive. Despite being marketed as aligned, even popular commercial LLMs like Claude, Grok, ChatGPT, and Gemini can provide harmful information when prompted. The US Government Accountability Office recently found that AI models like these could aid in planning a terror attack. Techniques like toxic training/fine-tuning (TTFT) and de-aligned training/fine-tuning (DTFT) can be used to create models with reduced or removed refusal behavior. This has legitimate uses, but also poses risks, as uncensored models can be used to provide guidance on illegal activities, assist in malware development, or generate harmful content. The study identified over 11,000 ULLMs across multiple platforms, with Hugging Face being the dominant upstream source. Only about 13% of the identified ULLMs carried moderation warnings, and security threat indicators were not consistently preserved across platforms, creating a supply chain problem.",
  "summary": "Making a mockery of alignment",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}