EVE CoreGuard is a deterministic decision-enforcement engine for regulated AI. Before an AI action executes, CoreGuard evaluates the proposed action against versioned policy packs and returns a decision — ALLOWED, BLOCKED, or MODIFIED — with no large language model anywhere in the decision path, so the same input always produces the same, replayable outcome. Every decision produces cryptographic evidence: an asymmetrically signed decision certificate — offline-verifiable by any auditor or regulator without contacting EVE (Ed25519, or KMS-backed ECDSA P-384 with the private key held in hardware). Certificates chain into a Merkle-rooted, tamper-evident audit trail with open-format export. CoreGuard governs single decisions and, on the Agent Governance tier, multi-step tool-using AI agents. It ships with standard policy packs, supports custom policy rules, and includes an SR 26-2 documentation package to support model-risk-management review. Built by EVE Core LLC, part of EVE NeuroSystems LLC.
