CoinFund founder Jake Brukhman said the Anthropic export-control dispute shows why decentralized AI networks are gaining attention as a possible counterweight to centralized model control.
Summary
- Jake Brukhman said Anthropic’s model shutdown shows how frontier AI can face direct government control.
- Decentralized AI teams are testing distributed GPU training as centralized compute access faces tighter rules.
- Pluralis is exploring tokenized AI model ownership by splitting weights among network participants and operators.
In a June 13 X post, Brukhman said AI models are a centralizing force and a major target for government control. He linked that view to Anthropic’s decision to comply with a U.S. directive that forced it to suspend access to Fable 5 and Mythos 5.
Anthropic said the order required it to block access for foreign nationals, including foreign-national employees, whether inside or outside the United States. The company said it disabled both models for all users to meet the directive, while other Claude models remained available.
Anthropic shutdown highlights distributed GPU training
Brukhman said decentralized networks can act as a counterbalance because AI’s first hard problem is access to large-scale compute. “The answer is simple: there is enough commodity GPU compute in the world to compete on the frontier,” he wrote, adding that new training methods are needed to use it.
He named Gensyn, Prime Intellect, Bagel, Pluralis, Nous Research, Macrocosmos AI, and Covenant as teams working on distributed training. He said their work shows that decentralized training is possible, cheaper, and nearly as efficient as centralized systems, though the sector still faces technical limits.
Pluralis model draws business focus
Brukhman also pointed to the business problem facing open-source AI. He argued that open models can be useful but often lack a revenue model strong enough to support frontier training costs.
He said Pluralis has proposed one answer by splitting model weights among participants. In his view, that structure can support tokenized AI models because no single participant holds the full model, while the network can still provide access to the system.
Broader AI pressure
As previously reported by crypto.news, Anthropic launched Fable 5 only days before the shutdown, presenting it as a Mythos-class model with added safeguards. The same report said some cybersecurity, biology, chemistry, and distillation requests would fall back to Claude Opus 4.8.
Earlier reports also showed the scale of AI infrastructure demand. Blackstone and Apollo were lining up about $36 billion in debt financing for Anthropic’s Google TPU expansion. Separate reporting on open AI infrastructure said concentrated compute access can leave whole regions dependent on a few providers.
Brukhman framed the moment as a choice between centralized AI and public AI on open networks. “This is the moment of truth,” he wrote, asking whether AI will fall under “censorship and unilateral government control” or move toward decentralized systems.
