PewDiePie just got banned from OpenAI. Twice. And instead of quietly moving on with his life like a normal person, he built his own AI model, stripped out the guardrails, and uploaded a video with a title that pretty much dares you not to click it: “I seriously should NOT be dropping this.” Classic Felix, honestly.

So What Actually Happened Here

The YouTuber (yes, still a YouTuber, even after all these years, even after the chess phase and the Minecraft phase and whatever else he’s been through) unveiled something called Ajax this week. It’s an “uncensored AI model,” built specifically to run on your home PC instead of living in some data center you’ll never see. In the video, he shows off an actual email from OpenAI explaining why his account got deactivated. The reason listed: distillation. For a closer look at how the uncensored home PC model actually runs, there's a full breakdown of Ajax's setup.

If you don’t know what that means, don’t worry, most people don’t. Distillation is basically when you take a big, expensive AI model, pull outputs or reasoning traces out of it, and use that data to train a smaller, cheaper model of your own. Think of it like… I don’t know, photocopying someone’s notes instead of going to the lecture yourself. OpenAI clearly doesn’t love when people do this to their models, and they’ve gone after other companies for it before. Now apparently they’re going after individual creators too.

And look, I get why OpenAI cares about this. Training these models costs an absurd amount of money. If random YouTubers can just skim the cream off the top and build competing products for free, that’s a real problem for their business model. But banning the guy twice? That’s not a slap on the wrist, that’s OpenAI basically saying “we see what you’re doing and we don’t want you anywhere near us.”

What Ajax Actually Is

Here’s the part that’s actually interesting to me, more than the ban drama itself. Ajax isn’t some standalone chatbot. It’s the engine behind something called Odysseus, which is PewDiePie’s self-hosted AI workspace. We’re talking an always-on, autonomous agent that’s supposed to handle your actual life stuff: searching the web, browsing, email, your calendar. All running locally, all “completely private,” according to the page he put up for it. That cost pressure is part of a bigger pattern, one lawmakers are now probing in the AI tax gap tied to Big Tech's AI spending.

Under the hood, it’s a fine-tuned version of Qwen3.5-9B, Alibaba’s model, not some from-scratch creation. That matters, because it means the “uncensored” part isn’t really about building new intelligence, it’s about ripping the safety training back out of an existing model. The page straight up says the refusal behavior has been removed “for a freer, less restricted AI experience,” then kind of sheepishly tells users to use it responsibly. Which, sure. Good luck with that.

Does Any Of This Actually Work?

This is where I’ll admit I’m skeptical, not gonna lie. The setup he’s going for sounds a lot like OpenClaw or Nous Research’s Hermes Agent, projects that have been chasing this exact dream for a while now: a private, local, always-on agent that doesn’t phone home to a corporation every time you ask it something. It’s a genuinely appealing idea! Nobody loves the feeling that their calendar and email are getting scraped by some company’s servers 24/7.

“I seriously should NOT be dropping this.”

But “uncensored” is doing a lot of heavy lifting in the marketing here. Removing refusals from a 9B parameter model running on a home PC isn’t the same thing as building some rogue superintelligence, despite how the framing might make it sound. It’s a smaller, fine-tuned model with the safety rails sanded off. That can genuinely be useful for people who are sick of models refusing harmless requests. It can also go badly in ways that are pretty predictable once you remove the stuff that was there specifically to catch edge cases.

The Bigger Fight Nobody’s Talking About

What’s actually going on here is bigger than one YouTuber and one banned account. This is the open-source-versus-closed-source AI fight playing out in real time, with a famous face attached to it instead of some anonymous research lab. OpenAI wants to protect what they built. Independent creators and researchers want the freedom to build on top of existing models without asking permission first. Both sides think they’re the reasonable one. Neither side is entirely wrong.

What makes this particular story land differently is the audience. Most distillation disputes happen between companies, lawyers in a room somewhere, nobody outside the industry even hears about it. PewDiePie has over 110 million subscribers. When he makes a video about getting banned for building an uncensored AI, that reaches people who’ve never thought about model distillation in their lives. Suddenly it’s not an abstract policy debate, it’s “that guy from YouTube got banned, let’s go see why.”

What This Actually Means

Honestly? I think this is a preview of where a lot of these fights are headed. As it gets easier for any person with a decent GPU to fine-tune an open model and strip out the guardrails, companies like OpenAI are going to keep drawing hard lines around what counts as acceptable use, and creators with big platforms are going to keep testing those lines, loudly, on camera, with millions of people watching. Meta's AI ad tool shows the same tension playing out on the corporate side, where guardrails and automation keep colliding.

Is Ajax actually going to replace your assistant, your calendar app, your whole digital life? Probably not, not yet anyway. But the fact that it exists, that it’s being built in public by someone with this much reach, tells you something about where the appetite is right now. People want AI that answers to them instead of a terms-of-service document. Whether that’s a good idea in practice is a different question entirely, and I’m honestly not sure anyone, including PewDiePie, has fully answered it yet.