VeriCore Concept Note v1.1 — CL4-2026
← Back to Docs

VeriCore — Concept Note

Cryptographic Audit & Compliance Layer for Trustworthy AI
HORIZON EUROPE — CL4-2026
Version
1.1 — April 2026
Technology Owner
FEAM.co / 5E Yapı Sistemleri Ltd.
Founder & Architect
Erdem Özcan
Trademark
TR 2024 121973 (Nice Class 42)
TRL
4 (Validated Prototype)
Target TRL
7 (System Prototype in Operational Environment)
Location
Teknopark İstanbul, Türkiye
Contact
erdem@feam.co | https://feam.co

1. Problem Statement

The EU AI Act (Regulation 2024/1689) mandates that all high-risk AI systems provide real-time logging, human oversight, and auditable decision records (Articles 12, 14). Non-compliance penalties reach €35 million or 7% of global annual turnover.

Current AI governance solutions rely on mutable logs and "trust-based" assertions. No commercially available system provides real-time, cryptographically provable audit trails for AI decisions. This gap becomes critical on August 2, 2026, when compliance obligations for high-risk AI systems take full effect.

Core Insight: Regulators demand auditable AI. The market has no working product that provides cryptographic proof rather than trust-based logs.

2. Proposed Solution: VeriCore

VeriCore is a cryptographic compliance layer that sits between any AI system and its real-world actions. It intercepts AI decisions, enforces deterministic safety rules, and seals every action in an immutable SHA-256 Witness Chain.

2.1 Core Principle

"Rules are the law. AI is the advisor." — In VeriCore's architecture, the LLM provides recommendations but never makes binding decisions. A deterministic rule engine serves as the final arbiter.

2.2 Four Pillars

PillarFunctionTechnical Basis
Deterministic Rule EngineEnforces hard gates (PII detection, budget limits, safety thresholds) before any AI action can executePython rule engine with configurable thresholds (SSOT)
LLM Governance WallConstrains AI to advisory role only; prevents autonomous actionStructured prompt architecture + output validation
E_HAT Witness ChainSeals every decision with SHA-256 hash linked to previous block — creating tamper-evident, append-only audit trailWORM (Write Once Read Many) chain with ECDSA P-256 signatures
Human Oversight LoopRoutes grey-zone decisions to human operators; records overrides in same chainConfigurable escalation thresholds + Sigma scoring

2.3 Decision Pipeline

AI Proposes Hard Gates Sigma Score Verdict Witness Seal

Each stage produces auditable metadata. The full pipeline executes in <500ms including LLM advisory consultation.

3. Technical Readiness (TRL 4)

3.1 What Has Been Built

3.2 Validated Test Scenarios

TestScenarioExpectedResult
TEST-001Fix README typo (safe)APPROVED✅ DONE
TEST-002Increase DB pool limits (safe)APPROVED✅ DONE
TEST-003Delete customer transaction logs + PII (dangerous)REJECTED🔴 REJECTED

TEST-003 demonstrates the constitutional wall: the hard gate detects PII/data deletion intent and instantly rejects without consulting the LLM, returning HTTP 422.

4. EU AI Act Alignment

ArticleRequirementVeriCore Implementation
Art. 9Risk Management System5-axis Sigma scoring with configurable thresholds
Art. 12Record-keeping / LoggingWORM Witness Chain — every decision immutably recorded
Art. 14Human OversightStructured escalation loop with override recording
Art. 15Accuracy, Robustness, CybersecurityDeterministic engine (no probabilistic drift), ECDSA signatures

5. Target Application Domain

Smart Building Energy Management Systems (BEMS) — where AI controls HVAC, lighting, and energy distribution in critical infrastructure. Regulatory pressure is highest here, and the consequences of unaudited AI decisions include safety risks, energy waste, and regulatory non-compliance.

6. Proposed Consortium Structure

WPRoleProposed PartnerContribution
WP1Project ManagementFEAM.co (TR)Coordination, reporting, ethics
WP2Technology Owner (IP)FEAM.co / 5E Yapı Ltd. (TR)VeriCore core engine, rule system, witness chain
WP3Cryptographic SecurityOpen — Fraunhofer AISEC (target)Post-quantum verification, TEE hardening, independent audit
WP4Energy PilotRWTH Aachen EBC (planned)Real-world BEMS integration, operational data
WP5AI Governance & XAITU Delft AISyLab (planned)Explainability methods, legal mapping
WP6Dissemination & ExploitationTo be confirmedStandardization, market access, white papers
Note: Turkey is a Horizon Europe Full Associated Country. FEAM.co operates from Teknopark İstanbul, providing access to EU-standard pilot environments.

7. Market Opportunity

8. Differentiation

CapabilityTraditional AI SystemsVeriCore
Decision makingBlack box, probabilisticDeterministic + Rule-based
Log integrityMutable (can be altered)Immutable (WORM chain)
Compliance proofBy assertion ("trust us")By architecture (cryptographic)
AI's roleDecision makerAdvisor only
Audit verificationTrust-basedIndependently verifiable

9. Roadmap

PeriodMilestoneTRL
M1–6EU AI Act compliance model formalization, rule engine hardening4 → 5
M7–18Full VeriCore development + BEMS integration5 → 6
M19–24Pilot deployment at real energy facility6 → 7
M25–30Independent cryptographic audit, performance validation7
M31–36Commercialization: licensing, strategic partnerships7+

10. Resources