The new installment of Chain of ReasonThe Brownstone Research newsletter written by Ben Lilly, says that the open source artificial intelligence battle is following the same path that Bitcoin took a decade ago, and that pattern-conscious investors are profiting.
The note opens with evidence that Anthropic CEO Dario Amodei gave to Congress in July 2023. Amodei admitted that open source is “a good thing” in many fields of science and that the risks of open models released so far are “limited,” but he warned that the measurement of open source models is moving “in a very dangerous way.”
Lilly clearly reads the subtext: if open models are dangerous, then closed models sold by companies like Anthropic are a safer option – and the next policy is to limit open and lift closed.
Bitcoin’s early critics are looking at what AI is up against
That framing is one that digital asset investors know well.
He also revisits Bitcoin’s early critics, from Rep. Jared Polis buying the first Bitcoin on Capitol Hill in 2014 to send a call to Sen. Joe Manchin to ban “dangerous money,” through 2023 allegations that regulators are trying to cut off crypto from the banking system in what critics call “Operation Choke Point.”
The industry survived, he notes, and Washington is now moving toward clearer regulations with the passed GENIUS Act and the pending CLARITY Act.
Artificial AI, which Lilly calls “DeAI,” has that fight now. He points to recent developments as evidence that the walls are going up: the US export ban on Anthropic’s latest release, which he says will push the company toward permissioned access that verifies a user’s identity before handing over a model, and OpenAI’s decision to limit its GPT-5.6 release to trusted partners.
He expects the identity requirements to spread. He writes: “It’s for your own protection. “It always is.”
The note relies on a national security anecdote to explain the fear driving these measures. Lilly quotes NSA chief Joshua Rudd, on Sen. Mark Warner, explaining how the Anthropic “Mythos” model has penetrated “virtually our entire isolated system, not in weeks, but in hours.”
However, open source fills the gap, according to the piece. Lilly says the latest GLM-5.2 scored on par with Anthropic’s Sonnet 4.6 from February, leaving open models about three to four months behind the threshold, and predicts an open challenger for Mythos and GPT-5.6 in the fall.
He says the bigger opening is decentralized training on peer-to-peer networks like Bitcoin and Ethereum – swapping compute-for-network-security for compute-for-model-training. Distributed training, he notes, has grown from less than one billion parameters to 100 billion in two years.
He names three startup projects – Dark Bloom, which enables low-cost privacy for idle Macs; c0mpute, a decentralized computing network; and Pluralis, which trains AI on all distributed consumer GPUs — and expects more to launch tokens and reward users for computing contributions.
The note concludes with the idea that governments will try to block open models and will fail. To him, investing in the area “would be like buying Bitcoin in 2014, back when it was still ‘risky.’
