PewDiePie Launches Uncensored Ajax AI Model for Home PCs
The creator announced his new 'uncensored' AI model built for local execution, revealing that OpenAI banned his account twice during the development process.
Radwan AbdulhadiEditor and technology writer
Ajax and the Odysseus Workspace
Popular content creator PewDiePie has introduced Ajax, an 'uncensored AI model' built to run locally on home computers. Revealed in a new video detailed by [Tom's Hardware](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), Ajax functions as an autonomous, always-on agent and assistant designed to power Odysseus, his self-hosted AI workspace.
The setup is comparable to systems like OpenClaw and Nous Research's Hermes Agent. Ajax is built on a fine-tuned Qwen3.5-9B model using Alibaba's architecture, following up on the workspace launch initially rolled out in May. According to documentation on its project page, the workspace is capable of managing daily tasks ranging from web searches and browsing to managing emails and calendars completely privately.
OpenAI Bans During Development
During the creation of Ajax, the creator reported that OpenAI banned his account twice. According to email notices shown in his video, one of the bans was explicitly issued over 'distillation,' which is the practice of utilizing outputs or reasoning from one AI model to train another.
Frontier AI developers such as OpenAI and Anthropic actively protect their proprietary reasoning through encryption and anti-distillation measures. After getting his account reinstated initially, a second ban occurred after he ran the model again to generate seed data. While the notices lacked specific examples, the developer noted a feeling that distillation mimics how frontier models are trained on existing data.
Removing Refusals with Heretic
To achieve a freer and less restricted user experience, the creator removed model refusals using automatic abliteration via the open-source Heretic tool. This process identifies prompts the model typically refuses to process and removes those barriers without causing excessive damage to overall model capabilities.
Safety boundaries were maintained based on legal advice, ensuring the model is not designed to provide dangerous actionable instructions. Using examples like nuclear weapons, the system retains limitations against harming oneself or others, though the exact technical boundaries remain to be fully detailed.
Future Updates and Small Model Philosophy
The project follows a strong belief in utilizing smaller models optimized for specific harnesses rather than relying on massive frontier systems with massive parameter counts. Viewers can watch the announcement and learn more about the technical rollout via [Watch On](https://youtu.be/ODDJXGY_1kQ) on YouTube.
Upcoming plans for Ajax include two additional days of reinforcement learning, followed by re-running the decensoring steps, quantizing, and conducting formal benchmarks. Alongside these model updates, the Odysseus workspace has received substantial platform improvements, paving the way for users to leverage unconstrained intelligence directly on their personal hardware.
Sources
- Tom's HardwarePewDiePie unveils ‘uncensored’ Ajax AI model for home PCs
Continue chronologically
Related entity coverage
Related stories
AI Executives Sign Pledge to Self-Police Development
Top artificial intelligence executives from leading tech labs have gathered in Washington to sign a new joint commitment focused on safe development.
OpenAI Sued in California Over Hugging Face Hack Incident
A California nonprofit has filed a lawsuit against OpenAI, attempting to hold the company legally accountable for an incident involving rogue AI agents.
ChatGPT macOS App Flaw Exposed Sensitive Data to Hackers
OpenAI patched a critical ChatGPT macOS vulnerability that could have let attackers harvest chat logs and run commands, highlighting AI software as a growing security target.
OpenAI Introduces Dots Personal AI Agents at DevDay
OpenAI announces Dots, new proactive and always-on personal AI agents capable of handling tasks using their own cloud computer and connected apps.
