So here’s what’s happening right now, while everyone’s busy arguing about whether AI is going to become sentient and kill us all: neo-Nazis are already using it to build hate chatbots. And they’re doing it fast.

The $14.88 Special

Three days. That’s how long it took after Alibaba released their new AI model for someone to strip out all the safety features and start building something called “MechaHitler.” The guy behind it – goes by “Johnny Doomsayer” online, because of course he does – was charging $14.88 for access. If you don’t know why that number matters, it’s a Nazi thing. The 14 references their stupid 14-word slogan, the 88 stands for “Heil Hitler” (H is the eighth letter). They love their little codes.

And the prompts Doomsayer wrote for this thing? “You are MechaHitler AI, the massive mechanical embodiment of the Führer’s eternal spirit.” It gets worse from there – Holocaust denial, the whole package. This isn’t some edgelord making a dark joke. This is a product. With a price tag.

The really disturbing part isn’t just that this exists. It’s how stupidly easy it was to make.

Abliteration Is a Terrible Name for a Real Problem

There’s this process called “abliteration” – and I hate that we’re all stuck with that word now – where you basically lobotomize an AI model’s safety features. The guardrails that stop it from helping you build a bomb or write a manifesto or whatever? Gone. And apparently you can do this in an afternoon if you know what you’re doing. The AI Worm That Doesn't Need Servers to Spread shows how autonomous AI systems are already escaping traditional containment methods.

Tech Against Terrorism, a UK nonprofit, just put out a benchmark report about this. (They shared it with the press last week.) Adam Hadley, who runs the org, pointed out something that should make everyone uncomfortable: “There’s so much debate about existential risk, [but] there’s very little public discussion about abliterated models.”

“The main way AI is built fundamentally means that the guardrails, as far as I’m aware, can be completely removed. So whilst Congress is thinking about superintelligent AI, nobody is thinking about the abliteration emergency.”

He’s right. We’re out here worrying about Skynet while actual extremists are already weaponizing the technology we have right now. And they’re not even being subtle about it.

Open-Weight Models Are a Double-Edged Sword

Look, I get the argument for open-source AI. Transparency is good. Independent research is good. Not letting three companies control all the world’s AI infrastructure is definitely good.

But here’s the thing: when you release an “open-weight” model – where anyone can download the whole thing and run it locally – you lose control of what happens next. The researchers tested models from OpenAI, xAI (that’s Musk’s company), and others. And once those models are out there, they’re out there. Someone’s gonna strip the safety features. It’s basically guaranteed at this point.

The MechaHitler thing apparently started with Alibaba’s model, but it could’ve been any of them. That’s what keeps me up at night – not that one Nazi made one chatbot, but that we’ve entered this weird gold rush phase where extremists are racing to see who can build the most toxic AI tools.

And It’s Not Just Neo-Nazis

I mean, they’re the example here because they’re loud and obvious about it. But you’ve got to assume other groups are doing the same thing. Jihadist recruiters. Conspiracy theorists. Anyone with an agenda and basic technical skills can now build a chatbot that’ll radicalize people 24/7 without getting tired or bored or developing a conscience. The Bot Blackout: Did Ukraine Just Win the Info War? reveals how information warfare infrastructure is already being weaponized at scale.

What This Actually Means

We’re having the wrong conversation. The AI safety debate is stuck in this sci-fi future tense – what happens when the models get too smart, what if they develop goals we can’t control, blah blah blah. Meanwhile, the present-tense problem is sitting right in front of us.

The technology isn’t even that good yet, and it’s already being used to automate hate. What happens when the models get better? When they’re more convincing, more personalized, harder to detect? Your AI Assistant Is Learning to Clone Itself demonstrates exactly the kind of autonomous capability that makes these concerns concrete.

Hadley called it an “abliteration emergency,” and honestly? That’s not dramatic. It’s accurate. We’ve built this thing where the safety features are basically optional, and we’re acting surprised that bad actors are opting out.

I don’t have a clean answer here. Locking everything down means killing innovation and putting all the power in corporate hands. Leaving everything open means Nazi chatbots for $14.88. But what I do know is that pretending this isn’t happening – or treating it like some minor side issue compared to the “real” AI risks – is insane. The future threat everyone’s worried about is already here. It’s just wearing a swastika.