PewDiePie built his own AI model then OpenAI shut his account

Felix Kjellberg, better known as PewDiePie, has announced Ajax, the first AI model of his own. It's a fine-tuned version of Alibaba's Qwen 3.5 9B that powers Odysseus, the free, open-source ChatGPT alternative he launched in May, handling web search, browsing, email and calendar on local hardware. He used chain-of-thought outputs from OpenAI's Sol model to help train it, and OpenAI deactivated his account over "Distillation". He says it was restored after an appeal, then suspended again.

The route he took to get here is strange. Minecraft videos, then building Arch Linux setups on camera, then a home lab of eight modified RTX 4090s he calls The Swarm. He picked a small model because fewer people could run a bigger one, and his stated aim is to let more people try AI on their own computers. Early testing showed roughly 90% task completion, and he calls it a work in progress. As of 2 October there was no model card, no public benchmarks and no confirmed weights licence, so Ajax isn't really out yet.

Distillation means using a bigger model's answers to teach a smaller one. OpenAI's Terms of Use prohibit using its output to develop competing models. OpenAI hasn't published any explanation of the bans, so everything so far is his side of the story.

The hypocrisy argument has teeth. OpenAI trained its models on huge amounts of web content, books and code without asking the people who made them, and some of those people are suing. Then it says you can't train on its outputs. That looks like taking being fine when they do it and theft when it's done to them.

I'd still be careful with the stronger claims. As far as I know, nobody has shown that OpenAI distilled from a competitor's model. The charge that sticks is scraping the open web without consent, which is arguably worse in scale but a different act. And Felix agreed to the terms. Scraping public pages and breaking a contract you signed feel morally similar, but legally they differ, and enforcing a terms clause against a customer is fairly standard.

The bans probably weren't personal either. The same week, OpenAI said it had disrupted a distillation campaign by people linked to China's Moonshot AI. Nobody has reported a link to Felix. My guess is his usage tripped detection built for that kind of operation, and he got caught in the net.

So my view: calling out the double standard is fair, calling the ban unjust is harder. The better argument is that the industry needs consistent rules on training data, and right now the big labs are writing those rules in their own favour. Felix probably knows this. OpenAI as the villain fits his "own your AI" pitch perfectly.

If your business runs on someone else's AI account, the duller lesson is to read the terms, because an automated system can switch you off before anyone explains why.