An ai system security decision and control system and method

Abstract

This invention discloses a security decision-making and control system and method for AI systems. It includes a TPM (Telematics Management Unit), a remote proof unit, a bus access control unit, a decision routing engine, an execution licensing module, a consensus network, and an auditing module. The TPM performs integrity measurements on the system software and AI model; the decision routing engine, within a secure execution environment protected by the TPM, routes AI decision commands according to policy rules; the execution licensing module generates execution licenses by binding command digests, model integrity digests, and PCR (Computational Pricing Value) reference values; the bus access control unit performs peripheral access control through a hardware arbitrator or trusted firmware, intercepting invalid commands; the consensus network votes on non-real-time governance decisions; and the auditing module asynchronously stores evidence. This invention balances real-time performance and security, and is suitable for highly autonomous scenarios such as autonomous driving and robotics.

Images