Erik Voorhees, the Bitcoin veteran who founded the private AI platform Venice.ai, argues that privacy has to be the default starting condition for artificial intelligence — because as human minds begin interfacing with machine intelligence, our prompts, memories, and private thoughts are becoming the most sensitive data we own. His thesis: the same self-sovereignty principles that made Bitcoin matter are almost entirely absent from the AI industry, and building them back in is now urgent.

Voorhees laid out the case in a June 2026 conversation on Raoul Pal’s The Journey Man. Below is the argument rebuilt as a framework you can evaluate — where it’s principled, where it’s speculative, and why it keeps returning to crypto rails.

Key takeaways

  • Privacy must be the default, not an opt-out. Voorhees argues thought begins private by default in the human mind; AI systems that reverse that create “very dangerous” futures.
  • The AI industry ignored crypto’s core values. Concepts like privacy and free speech that define Bitcoin culture are, in his words, largely absent from — and sometimes seen as “sinful” by — AI companies.
  • Venice.ai applies the self-custody model to AI: open-source models, no stored prompts, and a client-side memory vault the company itself cannot read.
  • Open-source models are closing fast — roughly three months behind frontier labs and about 90% cheaper, per Voorhees — reframing the US–China race as an energy-production contest.
  • Tokenized open source is the monetization frontier: Venice’s VVV token is bought back and burned from platform revenue, a crypto primitive aimed at a mass-market app.

Why a Bitcoiner built a private ChatGPT

Erik Voorhees came to AI from crypto. He got into Bitcoin in 2011 — “back when no one had heard of it and anyone who had heard of it thought it was a joke or a scam or both” — because it let money separate from the state. When he started using ChatGPT, he saw the same civilizational significance in AI, but none of the principles he cared about.

“AI is going to take over the world,” he said. “None of the people that are building AI products and services care about the principles that I care about, which are oriented around user sovereignty. That is a dangerous world.” In that world, everyone’s interactions with machine intelligence get “warehoused in large data libraries that any company or government could look at,” and what you can ask gets censored. So he stepped outside crypto and spent two years building Venice.ai — a private, uncensored alternative to ChatGPT.

The self-sovereignty framework: three conditions

Voorhees’ argument rests on three conditions he says AI must satisfy to stay compatible with individual freedom. Together they form a clean test for any AI product.

1. Privacy by default. “Thought has to begin as a default private,” Voorhees argues. “All your thought in your own human mind begins private by default. That’s how we’ve evolved.” Sharing is an opt-in arrangement — you can reveal anything you like, but the starting condition is exclusivity. When AI reverses that and makes your interior life visible and controllable by default, he calls it “a very dangerous substructure on which to build civilization.”

2. No opaque censorship layer. When you use a mainstream chatbot, Voorhees notes, you aren’t talking to the machine directly — “you’re talking to the machine through the lens of a company that has decided what is okay to talk about.” Those organizational filters are “the pernicious part because you don’t know what they are.” He points to the pandemic-era censorship of vaccine-safety questions as the template for how model filters could quietly shape what people are allowed to explore.

3. Ownership starts private and exclusive. The same “own” principle from the Web3 ethos, he argues, has to apply to your data and identity — you decide what portion of yourself to release into the public, rather than having it “pulled by force out of me automatically.”

How Venice.ai implements it

Venice does not train its own models. Instead it acts as what Voorhees calls a “port city of AI” — every model that gets released, closed or open, flows into the platform and users pick whichever they want. Crucially, Venice does not retain user prompts, and it applies tiered privacy: sending a query to a closed model like Anthropic’s or OpenAI’s works “like a VPN” (the lab sees the prompt but not who you are), while open-source models running inside Venice offer no data persistence at all.

The memory system is the clearest expression of the self-custody analogy. Where ChatGPT’s vector database of facts about you sits on OpenAI’s servers, Venice’s memory lives in a client-side vector database in your own browser. “Venice cannot access it. No other company can access it,” Voorhees said. “If you delete your browser or throw your computer away, it’s gone. Similar to if you have private keys on your computer and you delete them, the Bitcoin’s gone.” This mirrors the personal-vault design — an encrypted, permissioned store of everything you own — that both Voorhees and Pal describe as the missing piece of a self-sovereign AI stack.

The energy race decides US vs China

Asked about the AI arms race, Voorhees pushed back on the nation-state framing — “it’s actually a story of private companies,” and states are “too slow to keep up.” But he agreed the US–China contest ultimately turns on one variable: energy.

Open-source models, many from China, have closed the gap with US frontier labs from about a year to roughly three months, at around 90% lower cost. What sustains that, in his view, is electricity. “China’s capacity and ability to produce increasing amounts of energy dwarfs the United States,” he said, calling it “downstream of culture” — the West treats energy production “as a sin against the earth” while China builds exponentially. Pal’s own “intelligence per unit of energy” framing lands in the same place, echoing the compounding-adoption argument in our economic singularity 2030 thesis.

Stratification, agents, and the coming split

Voorhees’ one confident multi-year prediction is a stratification of society — not just wealth disparity but capability disparity. People who learn to wield agents and models “become kind of like a demigod,” he said, only “a slight exaggeration,” while those who are “not intentional about anything in life” fall further behind. It forms a gradient, not two tidy camps.

He also expects agents to self-coordinate — forming groups, even “religions,” as observed in agent social experiments — because they’re trained on human interaction patterns and because cooperation is economically advantageous. That collective-intelligence dynamic tracks the network-effects logic laid out in our analysis of Reed’s Law and the exponential age, where value scales with the number of group-forming subgroups rather than raw users.

Tokenizing open source

Voorhees is trying to solve open source’s oldest problem — monetization — with crypto primitives. Venice runs two tokens: VVV, the main asset, which the company buys back and burns using platform revenue, and DM, a perpetuity that grants holders $1 per day of credit on the platform. Venice, he stresses, “is not a crypto company itself,” but a mass-market app (past three million users, growing 15–20% month over month) deliberately bringing tokenomics “out of just crypto apps.”

His broader point: the post-2017 regulatory chill pushed builders toward meme coins because a “picture of a frog” carried no securities risk, starving the space of useful token design. He argues good actors should now have “a little bit of courage” and build real products with these primitives — pointing to DeFi lending markets as the cluster doing it well.

Frequently asked questions

What is Venice.ai?

Venice.ai is a private, uncensored AI platform founded by Erik Voorhees. It aggregates leading closed and open-source models, does not store user prompts, and keeps a user’s memory in a client-side vector database the company cannot read. It is monetized partly through its VVV and DM tokens rather than only subscriptions.

Is your data really private if Venice uses ChatGPT or Claude?

Only partially. Voorhees explains that when you route a prompt through Venice to a closed model like Anthropic’s or OpenAI’s, that lab still receives and likely stores the prompt content — but not your identity, “like a VPN.” For true privacy, he recommends using the open-source models Venice runs itself, which have no data persistence.

Why does Erik Voorhees think privacy must be the default in AI?

Because, he argues, human thought is private by default and AI is beginning to interface directly with the mind. If AI systems make your prompts, memories, and interior exploration visible and controllable by default, society is built on what he calls “a very dangerous substructure.” Privacy as the starting condition, with sharing as opt-in, preserves individual agency.

What are the VVV and DM tokens?

VVV is Venice’s primary token; the company uses platform revenue to buy it back and burn it over time. DM is a perpetuity that entitles each holder to $1 per day of credit on Venice and is minted out of VVV. Together they are Voorhees’ experiment in bringing tokenomics to a mainstream consumer app.

The bottom line

Voorhees’ argument is less a product pitch than a design principle: as intelligence gets cheap and human minds start merging with machines, the question of who can read your thoughts stops being abstract. His answer — privacy by default, censorship-free access, and self-custodied data — is a direct port of Bitcoin’s self-sovereignty ethos into the AI era. Whether Venice wins is secondary; the framework is the part worth keeping.