Solana beta lets developers verify hardware behind private AI

The toolkit pairs a software development kit with an on-chain verification program developed by Nito in coordination with the Foundation. Intel TDX and AMD SEV-SNP attestations are supported. NVIDIA GPU verification is still in development, and the Foundation says the release has not been audited and is intended for developers building and testing applications.

In the documented design, a trusted execution environment isolates code and data from the host operating system and its operator. A confidential virtual machine creates a signing key that stays within the protected hardware. A hardware-signed report links that key to a recent Solana slot; the verification program checks the report and records the key, hardware type and expiration date. The machine signs its AI output with the key, allowing an application to check the signature and consult the on-chain record. The AI computation itself happens off-chain, and users’ conversations and inputs do not need to be published.

Solana’s chat template, built with Next.js and React, demonstrates the process. The browser checks the confidential environment before encrypting a prompt to its key; if verification fails, the prompt is not sent. The encrypted message passes through the application server and Nito to the protected environment, where the model generates a reply that is encrypted before returning to the browser. According to the template documentation, neither intermediary can read message contents. However, the application server can see a visitor’s IP address, while Nito can see the account, model, request timing and encrypted message size.

The design also depends on the webpage code developers serve, the manufacturer’s hardware attestation and the software running inside the environment. Verification identifies a published software release but does not show that it is free of bugs. Developers must also implement authentication, rate limiting and bot protection; the example requires a funded Nito account and API key.

The Foundation has cited personal assistants, agents using confidential business data and private trading strategies as possible applications. Other suggested uses include prediction-market or insurance rulings, trading orders accepted only from verified machines, and AI governance decisions. Nito is incorporating the infrastructure into its private AI platform, while other teams can use the open-source components.