Bittensor’s decentralized AI network is building a confidential routing layer that could reshape how AI inference gets priced, delivered, and verified. The integration centers on Subnet 28, internally called “gm,” which uses Trusted Execution Environments to let miners compete on routing costs, payments, and model access, all while keeping user queries private.
What Bittensor is actually building
Subnet 28 introduces a privacy-first marketplace for AI inference. Trusted Execution Environments, or TEEs, are hardware-level secure enclaves that process data without exposing it to the machine’s operator. Even the miners running your AI queries can’t see what you’re asking.
The system generates verifiable attestations, cryptographic proof that your request was handled inside a secure enclave without tampering. Miners on the subnet compete not just on speed or quality but on routing costs and payment terms, creating a dynamic pricing layer that sits on top of Bittensor’s existing subnet economy.
Development discussions around this confidential routing integration took place between May 28 and 29, 2026. The architecture has been explicitly compared to “OpenRouter or Venice AI, but on Bittensor,” suggesting the team sees centralized inference aggregators as the competitive benchmark to beat.
