NEAR AI has introduced a new infrastructure that ensures private AI model execution by using hardware-based confidential computing, protecting user data from exposure even to the service providers themselves. The platform leverages Intel TDX and NVIDIA's confidential computing technologies to run AI models within isolated environments known as Trusted Execution Environments (TEEs).
How NEAR AI’s Confidential Computing Works
The technology creates encrypted memory spaces where the AI code and data remain shielded from the host system, cloud providers, and NEAR AI. Users can verify the integrity of the execution through cryptographic attestation, confirming that models run untampered in genuine secure environments. This approach supports open-weight AI models like Llama or Mistral, which can be loaded directly into a TEE, as well as "proxied closed models" from providers that keep their weights private.
The core of this system is IronClaw 1.0, an open-source AI agent runtime unveiled in late July 2026. It operates autonomously within hardware enclaves where operators have zero access, and it already leads in industry benchmarks such as PinchBench and ClawBench.
Enterprise Focus and Partnerships
NEAR AI's efforts signal a focus on enterprise adoption, highlighted by its integration with Venice.ai in March 2026. This partnership enables fully private inference for text and image processing, ensuring user data and outputs remain shielded throughout the workflow. also NEAR AI became part of NVIDIA’s Inception program earlier this year, a move that aims to enhance its confidential computing for enterprise-grade applications.
As privacy continues to be a major hurdle for banks and organizations exploring blockchain and AI, platforms like NEAR AI’s confidential computing could be a big deal in securing sensitive data during AI model interactions.
This content is for informational purposes and does not constitute financial advice.
