{
  "id": 11420548,
  "title": "PewDiePie unveils ‘uncensored’ Ajax AI model for home PCs",
  "url": "https://urgent.news/2026/10/02/pewdiepie-unveils-uncensored-ajax-ai-model-for-home-pcs",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-02T10:30:00.000Z",
  "source": {
    "name": "Tom's Hardware",
    "slug": "tom-s-hardware",
    "url": "https://www.tomshardware.com/tech-industry/artificial-intelligence/pewdiepie-unveils-uncensored-ajax-ai-model-built-to-run-on-home-pcs-creator-says-openai-banned-him-twice-while-making-it"
  },
  "original_language": "en",
  "account": "Minecraft sensation PewDiePie has introduced Ajax, his \"uncensored AI model\" designed for use on home computers, in a recent video. OpenAI penalized him twice during Ajax's development, a fact he reveals in an email screenshot. One ban was for \"distillation,\" a technique where one model's outputs or reasoning is used to train another. Ajax is built within Odysseus, PewDiePie's self-hosted AI workspace and assistant, akin to OpenClaw and Nous Research's Hermes Agent.\n\nAjax, a refined Qwen3.5-9B model, empowers the workspace to execute tasks like web browsing, emailing, and calendar management entirely privately. The platform emphasizes a \"freer, less restricted AI experience\" and cautions users to use it responsibly. PewDiePie unveiled the initial version, powered by Alibaba's model, as a follow-up to his May workspace launch.\n\nIn response to the bans, PewDiePie stated his intent was to \"distill just a little bit\" to enhance his model. Major AI companies like OpenAI and Anthropic often obscure their reasoning through encryption and distillation methods, with OpenAI recently facing accusations of wrongdoing linked to Moonshot AI. PewDiePie was reinstated once, but following the second ban, he questioned how OpenAI could have known about his actions. He notes that some distillation might be comparable to how AI models are typically trained.\n\nTo uncensor Ajax, he employed automatic abliteration via the open-source Heretic, which identifies prompts the model refuses and removes the denials without causing significant \"brain damage.\" He refrains from generating harmful or dangerous instructions, drawing the line at causing harm to others or oneself. His lawyer advised him that the model is \"not designed to provide dangerous actionable instructions.\" For instance, PewDiePie uses nuclear weapons as a cautionary example, though the actual boundaries remain unclear.\n\nPewDiePie strongly advocates for small models trained for specific purposes as an alternative to massive \"one trillion parameter\" models. He believes such smaller models make more sense for everyday tasks. He remains focused on research and creating a usable model, emphasizing the importance of keeping technology accessible and uncensored on individual hardware.",
  "summary": "PewDiePie says OpenAI banned him twice while he made Ajax, a fine-tuned Qwen3.5-9B agent.",
  "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."
}