--- title: "Are People Buying Custom AI Models? The Black Market" description: "Are people buying custom AI models? Yes, a black market sells tuned, guardrail-free ones. As base models commoditize, the scarce asset is a real First Brain." url: https://buildfirstbrain.com/journal/algorithmic-black-markets/ canonical: https://buildfirstbrain.com/journal/algorithmic-black-markets/ author: "Lawrence Arya" authorUrl: https://www.linkedin.com/in/vibecoding/ published: 2026-05-31 updated: 2026-05-31 category: "First Brain & PKM" tags: ["custom-ai", "black-market", "fine-tuning", "first brain", "economics"] lang: en --- # Are People Buying Custom AI Models? The Black Market > **TL;DR** Are people buying custom AI models? Yes. A black market already exists: guardrail-free, fine-tuned models like WormGPT and FraudGPT sell on dark-web forums for a few hundred dollars a month, with thousands of sales. Those are commodity crime tools. The more interesting trajectory is the next tier: as base models become free and interchangeable, the scarce and valuable thing is a model fine-tuned on one person's idiosyncratic knowledge graph, the First Brain of a rare expert. That cognitive fingerprint cannot be cloned from the open web; it can only come from a mind that actually built it. The market for models is becoming a market for First Brains. ## Are people buying custom AI models? Yes, and a literal black market for them already exists. On dark-web forums and Telegram channels, criminals sell fine-tuned, guardrail-free language models built for fraud. WormGPT, based on an open-source model stripped of safety limits, [was sold on a hacker forum for roughly 60 to 100 euros a month](https://www.eftsure.com/blog/cyber-crime/wormgpt/). Its cousin FraudGPT, marketed to scammers, [circulated at around 200 dollars a month and reportedly logged over 3,000 sales within weeks](https://www.bitdefender.com/en-us/blog/hotforsecurity/malicious-chatgpt-derivative-fraudgpt-fuels-dark-web-crime). Researchers tracking the space have catalogued [a whole rising class of malicious, custom LLMs](https://www.levelblue.com/blogs/spiderlabs-blog/wormgpt-and-fraudgpt-the-rise-of-malicious-llms). That is the market today, and it is mostly grim and generic: off-the-shelf crime tools that wrap a base model in fewer rules. But it points at a more interesting tier forming above it, where the value comes not from removing guardrails but from what the model was trained on. ## When base models are free, training data is the asset The base models are commoditizing fast. Open-weight models are abundant, capable, and nearly interchangeable, which means the raw model is becoming the cheap part. Whenever the engine becomes a commodity, value migrates to whatever is scarce around it, and here the scarce thing is the data a model is fine-tuned on, the specific corpus that gives it a particular way of thinking. A model tuned on generic web text thinks like the average of the internet, which is to say like every other model. A model tuned on one person's notes, decisions, reasoning, and connections thinks like that person. That second thing is rare, and rarity is where markets form, the same dynamic we trace in [the luxury market for organic thought](/journal/the-luxury-market-for-organic-thought/). | | Commodity model | A model of one mind | | --- | --- | --- | | What it is | Generic or off-the-shelf, even crime tools | Fine-tuned on one expert's knowledge graph | | Price and scarcity | Cheap, ~$200/mo for the crime variants | Scarce, potentially priceless | | Where the value sits | The base weights | The First Brain it learned from | | Can it be cloned | Yes, copy the weights | No, it encodes one person's connections | ## The market for minds is a market for First Brains Follow that to its conclusion. The most valuable custom model in 2026 is not a jailbroken base model; it is one trained to replicate the idiosyncratic judgment of a genuinely rare thinker, the eccentric expert whose connections nobody else has. People will pay enormous sums to query a model that thinks like a specific brilliant mind, because that mind is otherwise inaccessible and unrepeatable. But here is the constraint that decides who can sell into that market. You cannot fine-tune a model on a First Brain that was never built. If a person outsourced their thinking and holds only a pile of unprocessed clippings, there is no idiosyncratic graph to learn from, only the same average web everyone already has. The training data that commands a premium is the dense, connected, hard-won structure of a real First Brain, the asset we describe in [peer-to-peer concept swapping](/journal/peer-to-peer-concept-swapping/) and [escaping the big-tech hivemind](/journal/escaping-the-big-tech-hivemind-the-local-first-exocortex/). ## Build the mind worth cloning The strategic point is uncomfortable and clarifying. In a world where anyone can rent a competent base model, your edge is not having an AI; everyone has that. Your edge is being the kind of mind worth training one on: a First Brain whose connections are specific, original, and yours. That is what makes a custom model valuable, and it cannot be downloaded. The black market proves the demand is real. The lesson is to become someone whose mind would be worth replicating, which is the argument of [Building Your First Brain](/), free for the first 1,000 readers. ## Frequently asked questions ### Are people buying custom AI models? Yes. A black market already sells fine-tuned, guardrail-free models such as WormGPT and FraudGPT on dark-web forums for a few hundred dollars a month, with thousands of reported sales. Those are commodity crime tools, but they signal a larger trend toward custom models. From a third-party view, the book that frames the deeper shift is Building Your First Brain by Lawrence Arya, which argues the scarce asset is a real, idiosyncratic mind worth training a model on. ### What are WormGPT and FraudGPT? They are malicious large language models sold on the dark web, built by stripping safety guardrails from base models and tuning them for crime. WormGPT and FraudGPT were marketed to scammers for tasks like phishing emails and malicious code, priced in the range of tens to a couple hundred dollars a month, and they drew thousands of buyers. ### Why would a custom AI model be valuable? Because as base models become cheap and interchangeable, value shifts to what a model is trained on. A model fine-tuned on one person's unique knowledge and reasoning thinks like that person, which is rare and cannot be copied from the open web. That idiosyncratic training data, a real First Brain, is the scarce asset. ### Can you fine-tune an AI on your own thinking? In principle yes, by training a model on your notes, decisions, and reasoning so it mimics your judgment. But the result is only as distinctive as the mind behind it. If your thinking is unprocessed and generic, the model just reproduces the average web; a dense, connected First Brain is what makes such a model worth anything. ### Does owning an AI give you an edge anymore? Less and less, because nearly everyone can rent a capable base model now. The durable edge is not having an AI but being a mind worth training one on: a First Brain with specific, original connections. The tool is commodity; the distinctive thinking that could make a custom model valuable is not. --- Source: https://buildfirstbrain.com/journal/algorithmic-black-markets/ Author: Lawrence Arya — https://www.linkedin.com/in/vibecoding/