Grayscale names Bittensor as decentralized AI pick

Zach Pandl, head of research at Grayscale, highlighted Bittensor (TAO) in a report on decentralized AI the firm published this week. The TAO token rose nearly 30% over a five-day span after the report and regulatory actions affecting Anthropic.

Pandl described Bittensor as a permissionless, global network that lets contributors run nodes, train models and share results across competing subnets. The protocol uses the TAO token to reward nodes that the network deems most useful. In a blog post, Pandl wrote: “Think of it as Bitcoin for AI: What Bitcoin did for digital money, Bittensor hopes to do for AI.”

Anthropic last week lost the ability to provide its Fable 5 and Mythos 5 models to foreign nationals under a U.S. export-control directive. The company temporarily disabled the models for all customers while it assesses compliance. U.S. officials cited a potential narrow jailbreak technique that could expose sensitive capabilities; Anthropic has argued the vulnerability is minor and already present in other public frontier models.

Separately, Anthropic is defending a class-action lawsuit in U.S. federal court. The complaint alleges the company imposed unclear and restrictive usage limits on commercial Claude products that affected paying customers, claiming stated tiers did not correspond to the access or performance advertised.

Pandl framed these developments as an example of concentrated regulatory and operational risk for centralized AI providers, and he presented decentralized networks as an alternative that spreads work and control among many independent participants across jurisdictions. He wrote that decentralized designs can reduce the impact of a single government order or court action on global availability.

Bittensor’s technical design uses subnets that compete to provide machine learning, agent services and inference. Nodes are evaluated by the network and receive TAO rewards based on performance. The protocol aims to operate without a central authority controlling access to models.

Grayscale’s report noted that institutional investors are reassessing exposure to centralized AI amid regulatory and legal developments. Pandl wrote he expects regulatory pressure on centralized providers to persist and suggested decentralized networks could attract capital if compliance requirements reshape industry flows.

Decentralized AI projects use blockchain or similar infrastructure to distribute training, inference and model ownership across many participants instead of centralized data centers run by a single company. Supporters say this can lower single-point-of-failure risk; critics point to challenges with coordination, quality control and scalability.