Frequently Asked Questions

EVE AI Core is deterministic AI governance enforcement infrastructure. It sits between your AI models and production traffic, enforcing safety policies through deterministic trust layers, cryptographic audit trails, and sub-millisecond veto decisions. Unlike probabilistic guardrails or prompt filters, EVE Core uses a Three-Layer Trust Infrastructure with no signal path from Execution back to Control, so the AI systems it governs cannot bypass it. Every enforcement decision is logged to a tamper-evident hash chain for regulatory audit.

The Three-Layer Trust Infrastructure separates AI operations into three isolated layers:

• Authority Resolution Layer: Deterministic policy enforcement. Evaluates every action against charter rules, CRD scores, and cognitive locks. Runs in hardware isolation with no writable path from the Governed Inference Layer.
• Governed Inference Layer: Model inference and response generation. The AI system operates here but cannot read or modify Authority Resolution Layer state.
• Cryptographic Authority Chain Layer: Cryptographic audit chain. Every governance decision, veto, and policy evaluation is recorded to an append-only, SHA-256 hash-chained ledger for regulatory compliance.

The critical design property: there is no signal path from Execution back to Control. The AI cannot modify the rules that govern it.

Guardrails and prompt filters are probabilistic -- they use pattern matching or secondary LLM calls that can be circumvented by adversarial inputs. EVE AI Core is fundamentally different:

• Deterministic, not probabilistic: The veto engine uses pure functions with zero I/O, zero threading, and zero global state. The same input under the same policy version produces the same enforcement decision.
• FPGA-compilable isolation: The Authority Resolution Layer is designed for FPGA/firmware compilation. There is no software path for the AI to modify its own governance rules.
• Pre-action enforcement: Policies are evaluated before actions are dispatched, not post-hoc filtered.
• Cryptographic evidence: Every decision is hash-chained to an immutable audit trail, providing legal-grade proof of compliance.

CRD (Confidence-Reality Divergence) is a calibrated metric that measures the gap between an AI system's stated confidence and verified ground truth. CRD scoring detects when a model is confidently wrong -- the most dangerous failure mode in production AI. The system maintains a Brier-scored calibration history and adjusts assertion thresholds per domain (factual, creative, medical, legal). High CRD triggers automatic response gating: hedging qualifiers for moderate divergence, full abstention for critical divergence.

The Truth Store is a verified fact database used for real-time contradiction detection. It maintains curated facts across multiple categories with confidence scores and source attribution. Every AI-generated response is validated against the Truth Store before delivery. When a response contradicts a verified fact, the system performs sentence-level correction automatically. The Truth Store addresses the critical LLM failure mode where training data cutoffs cause confidently wrong answers about current facts (e.g., outdated world leaders, superseded regulations).

AEGIS (Automated Epistemic Governance and Integrity System) is EVE AI Core's automated adversarial self-hardening framework. AEGIS continuously generates adversarial challenges against the governance pipeline -- prompt injection, social engineering, policy circumvention, and escalation attacks -- then measures whether enforcement holds. When a challenge reveals a weakness, AEGIS proposes policy amendments through the same governance approval chain (risk assessment, charter compliance check, drift budget validation) that governs all system modifications. This creates a closed-loop security posture that strengthens over time rather than degrading.

The circuit breaker is a global AI kill switch that halts all Execution Plane operations immediately when critical safety thresholds are breached. It monitors six indicators: improvement velocity, capability growth rate, recursion depth, autonomy drift, goal drift, and resource consumption. If two or more indicators reach CRITICAL simultaneously, the circuit breaker triggers an emergency shutdown -- stopping all model inference, freezing the Evidence Plane for forensic review, and requiring explicit human authorization to resume operations. The circuit breaker operates in the Control Plane and cannot be disabled by the AI systems it governs.

Sub-millisecond. The veto engine is a pure deterministic module with zero I/O, zero threading, and zero external dependencies. Charter compliance checks, cognitive lock evaluations, and drift budget validations execute in under 1ms. This is achieved through pre-compiled rule sets, frozen constant tables (15 charter rules, 5 ethical red lines, 13 protected invariants, 6 lock thresholds), and a design that targets FPGA firmware compilation for production deployments. Governance adds negligible latency to the AI inference pipeline.

No. The Three-Layer Trust Infrastructure provides hardware-level isolation between the Control Plane and the Execution Plane. There is no signal path from Execution to Control -- the AI system cannot read, modify, or circumvent its own governance rules. Hard-block vetoes (covering identity erasure, value corruption, safety violations, and 5 ethical red lines) are enforced deterministically and cannot be overridden by any software path. The veto engine is designed as a pure function module suitable for FPGA firmware, with a published C header defining the hardware API contract. Prompt injection, jailbreaking, and social engineering are structurally neutralized because the enforcement logic never runs in the same execution context as the AI model.

EVE AI Core provides compliance infrastructure for three major regulatory frameworks:

• EU AI Act: Article 9 risk management (continuous risk assessment via CRD scoring), Article 14 human oversight (circuit breaker and approval workflows), Article 12 record-keeping (Cryptographic Authority Chain Layer), and Article 15 transparency (deterministic audit trails).
• NIST AI RMF: Full coverage of Govern, Map, Measure, and Manage functions through the charter enforcement system, CRD calibration, resilience scoring, and the Three-Layer Trust Infrastructure.
• ISO 42001: AI Management System requirements addressed through the governance pipeline, cryptographic certificates, and tamper-evident audit chains that satisfy external audit requirements.

All enforcement decisions are recorded to the Cryptographic Authority Chain Layer with SHA-256 hash chains and Ed25519-signed certificates, providing legal-grade proof of compliance for auditors.

Every governance decision is recorded to an append-only Evidence Plane using SHA-256 hash chains. Each entry links to the previous via cryptographic hash, making retroactive tampering detectable. Auditors can verify chain integrity at any time through the API, and individual enforcement decisions can be exported as Ed25519-signed certificates. The audit trail covers charter compliance checks, veto decisions, drift budget consumption, CRD score evaluations, and circuit breaker activations. Retention is 7+ years for regulatory compliance.

EVE AI Core has 90 filed U.S. provisional patent applications filed, covering the Three-Layer Trust Infrastructure, CRD scoring, deterministic veto enforcement, FPGA-targeted governance isolation, AEGIS adversarial self-hardening, the circuit breaker mechanism, somatic marker decision weighting, state continuity enforcement, recursive autonomy governance, and cryptographic truth verification. Serial numbers range from 63/988,235 through 64/047,284. All production licenses include patent rights. See the full patent portfolio.

EVE AI Core is licensed as governance infrastructure with three tiers:

• EVE CoreGuard Enterprise (from $125K/year): Deterministic veto engine, CRD scoring, cryptographic audit chain, charter compliance API, and patent license for all 90 filed U.S. provisional applications. Designed for organizations adding governance to existing AI deployments.
• Enterprise Governance (custom-quoted): Full Three-Layer Trust Infrastructure deployment, AEGIS adversarial self-hardening, circuit breaker, Truth Store integration, multi-tenant isolation, federated trust exchange, and dedicated implementation support.
• Sovereign Infrastructure (custom-quoted): On-premise deployment with hardware enclave support (TPM 2.0 / Intel SGX / ARM TrustZone), custom charter rule authoring, FPGA firmware compilation, air-gapped operation, and regulatory pre-certification assistance.

A free Developer SDK is available for evaluation and integration testing (1,000 governed decisions/month). See full licensing details.

Evaluation is via a fixed-scope design-partner pilot. Your team validates enforcement against your own decisions and policies in a sandbox — no production data required — and the pilot fee is credited toward an annual enforcement license. Get started.

EVE AI Core supports three deployment models:

• Cloud-hosted (SaaS): Multi-tenant deployment with per-organization isolation, available on Enforcement and Enterprise licenses.
• Dedicated cloud: Single-tenant deployment in your cloud account (AWS, Azure, GCP), available on Enterprise and Sovereign licenses.
• On-premise: Air-gapped deployment with hardware enclave support, available exclusively on the Sovereign license. Includes FPGA firmware compilation for deterministic governance enforcement.