NewsMacroPewDiePie Says OpenAI Banned Him Twice While He Built Ajax, an Uncensored AI That Runs on Your PC

PewDiePie Says OpenAI Banned Him Twice While He Built Ajax, an Uncensored AI That Runs on Your PC

Author: Decrypt·

Key Takeaways

  • •PewDiePie says OpenAI banned his account twice during the building of Ajax, with the first ban reversed after he disputed it and the second occurring after he used GPT-5.6 Sol to create seed data, and OpenAI has given no public response.
  • •Ajax is reportedly a 9-billion-parameter model built on Alibaba's open-source Qwen 3.5 and fine-tuned to power Odysseus, the free, self-hosted AI app PewDiePie launched in June.
  • •To make Ajax uncensored, PewDiePie used the open-source tool Heretic to perform 'abliteration,' removing the model's built-in refusals while drawing the limit at requests causing harm to other people or to oneself.
  • •PewDiePie argues small local models offer cost and energy advantages, estimating that a rumored trillion-parameter frontier model would require about 27 of his computers and the electricity of roughly 150 houses.
  • •Ajax is not yet downloadable, as PewDiePie still needs to rerun the ablation, then quantize and benchmark the model before releasing it on his website.
PewDiePie Says OpenAI Banned Him Twice While He Built Ajax, an Uncensored AI That Runs on Your PC

PewDiePie said in a new video that OpenAI banned his account twice while he built Ajax, a small AI model meant to run on a home computer and browse the web on the user's behalf. Tech City Authority found no public response from OpenAI.

The YouTuber told the story in a new video on his YouTube channel. Ajax is what he calls "a fine-tuned AI model" — an existing model that received extra training for one specific job. Here, that job is working inside Odysseus, the free, self-hosted AI app PewDiePie launched in June.

Run on one's own machine, a model like this keeps emails and calendars at home instead of on a company's servers. There is also no subscription fee, though the user's own hardware does the heavy lifting.

Ajax is reportedly a 9-billion-parameter model built on Alibaba's open-source Qwen 3.5. Parameters are the adjustable numbers that make up an AI's "brain," and more is generally considered more powerful. That open-source base is what puts a project like this within reach of an individual builder: the starting model is public, so the fine-tuning happens outside any company's platform. The model has yet to be released, but it will be available on his website once PewDiePie thinks it is ready.

The two bans

PewDiePie said he wanted to distill from GPT-5.6 Sol, the flagship model OpenAI released on July 9. Distillation means training a smaller AI on a bigger one's answers, like a student copying the class genius's homework. The prize was Sol's reasoning tokens — the chunks of text a model writes on a hidden scratchpad while it works through a problem. OpenAI keeps that scratchpad encrypted. It is a valuable enough prize that labs guard it closely — and that others keep trying to pick the lock.

In the video, he said he found a study that cracks it using OpenAI's own API. He also said he would never have cracked OpenAI, but OpenAI banned him anyway. He called the ban unjustified, disputed it, and the account came back online some time later.

Then, he said, he used Sol again to create seed data — the starter examples a model learns from — and OpenAI banned him a second time. OpenAI's terms of use bar using its output to develop competing models.

The bans come against a broader backdrop. In August, researchers showed that encrypted reasoning blocks from OpenAI, Anthropic and Google could be replayed to a weaker sister model that wrote them out in plain text. They decoded 315, of them from public code repositories. On Sept. 30, OpenAI said it disrupted a campaign by people associated with Moonshot AI, the developer of Kimi, to extract its models' hidden reasoning. The company said it closed a pathway that let someone replay another user's encrypted reasoning to recover its contents.

Taking the "no" out of AI

To make Ajax uncensored — meaning it answers requests that ChatGPT-style assistants refuse — PewDiePie used Heretic, an open-source tool that removes a model's built-in refusals. It compares how the model reacts to harmful and harmless prompts, finds the pattern behind its refusals, and cuts it out.

This is known as "abliteration," essentially a technique that lobotomizes a model so it does not need to be jailbroken because it always agrees to do whatever it is asked to do.

The surgery is not clean. PewDiePie says Ajax took "a little brain damage," but that it handles tasks like sorting inboxes and browsing about nine times out of 10. He is also still training it with GRPO, a method where the model attempts the same task 16 times and learns from the attempts that work.

PewDiePie said he chose which refusals to remove and drew the limit at "causing harm to other people or causing harm to yourself." His lawyer, he said, told him to state that Ajax is not designed to give dangerous, actionable instructions. The arrangement shifts the guardrails: instead of a company's policy team deciding what the assistant will not discuss, the person running the model makes that call at home.

Small model, big pitch

His case for small models is cost and energy. By his own napkin math, one copy of a rumored trillion-parameter frontier model needs about 27 of his computers and the electricity of roughly 150 houses. Ajax sits at the opposite end of that math — a reported 9 billion parameters rather than a rumored trillion, sized to live on one machine with no monthly fee.

PewDiePie calls Ajax version one. He said he still has to rerun the ablation, then compress (quantize) and benchmark the model to check that nothing broke.

Ajax is not downloadable yet. At first, its download page carried a countdown ending at 08:25 Japan time on October 3. That timer no longer exists. Until the ablation rerun, quantization and benchmarks are done and he considers the model ready, his website is the place to watch.