Solana Foundation and Nito Release a Toolkit That Verifies Private AI Hardware on Solana

The Solana Foundation has released Solana Confidential Compute, an open-source toolkit built with Nito that lets developers build private artificial intelligence (AI) applications and verify the hardware they run on. The beta runs on Solana's developer test network and has not been audited. It includes a software development kit (SDK) and an on-chain program that checks hardware attestations, according to Solana Compass.

The design depends on trusted execution environments (TEEs): isolated areas of a processor that sign a report describing the code running inside them. The Solana program verifies that signed report, so an application can show on-chain that a model is running inside a sealed enclave and not on an ordinary server. The beta supports Intel's TDX and AMD's SEV-SNP enclaves. Support for Nvidia graphics processors, where most AI inference actually runs, is still in development.

The documentation is direct about the limits. The blockchain verifies the receipt but still relies on the hardware that issued it. The application server can still see users' IP addresses. Nito can observe the account, the model, the timing of requests and the size of encrypted messages. Developers must add their own authentication, rate limiting and bot protection.

The launch adds a hardware-based option to Solana's existing privacy tools. Those include Confidential Balances, an encrypted-balance token extension with an optional auditor key, and multiparty-computation networks such as Arcium. The demand side was the subject of a Bitcoin Magazine interview with Mark Suman, chief executive of Maple AI, who described private AI built so that even its own team cannot read users' chats.