EU AI Act — Articles 14 + 15 (Human Oversight + Accuracy/Robustness).
Issuer: European Union — Regulation (EU) 2024/1689 · Year: 2024 (high-risk obligations effective 2026) · Source: official text →
Scope: High-risk AI systems placed on the EU market or used in the EU — including AI in critical infrastructure, education, employment, credit-scoring, law enforcement, justice, migration, and biometric identification.
What KYE Protocol™ supplies
Articles 14 + 15 require structurally-enforceable human oversight, accuracy + robustness measures, and full transparency on high-risk AI systems. KYE Protocol™ supplies the GovernedUI™ approval rail (Article 14) + the Decision Map™ + audit-chain replay surface (Article 15) that turn these articles from outcome-mandates into structurally-enforceable runtime guarantees.
Per-clause control mapping
| EU AI Act clause | KYE Protocol™ binding |
|---|---|
| Art. 14(1) — Human oversight: measures | GovernedUI Action Approval — every high-risk action passes through a structurally-enforced approval (none / single / two-person / two-person-with-legal modes). |
| Art. 14(4)(a) — Properly trained natural persons | Operating Model + Authority Scope — operators are entity-graph leaves with bounded purpose + authority. |
| Art. 14(4)(b) — Awareness of automation bias | Critical Point Review module — diff view + counterfactual replay surface for every decision. |
| Art. 14(4)(c) — Correctly interpret outputs | Decision Map™ — every decision binds the input, the rules applied, the signals fused, and the verdict; replayable offline. |
| Art. 14(4)(d) — Decide not to use or override | Two-person-with-legal approval mode + KYE Meta-Governance Gate™. The structural property that prevents self-grant is part of the patent track and is not disclosed in this repository. |
| Art. 14(4)(e) — Halt the system | Kill-switch endpoints on every agent + edge-governance compiled authority bundle. |
| Art. 15(1) — Accuracy, robustness, cybersecurity | Conformance Pack 133 fixtures + drift cascade + Reconciliation Engine bijection — every claim is testable. |
| Art. 15(4) — Resilience against errors | Reality Coupling rail + stable-drift detector — silent compromise + reality drift surfaced as signed events. |
Every binding above resolves to a canonical KYE Protocol™ artefact (engine, schema, audit event, or patent claim). The full per-control register is published in the conformance repo at github.com/KYE-Protocol/app/tree/main/internal.
What an auditor / regulator gets
- Replay Proof™ — re-derive any decision offline using only the publisher’s published JWKS. No back-channel to KYE™ project.
- Evidence Pack™ — sealed, signed, replayable container of decisions + bound rules + audit-chain anchors.
- Conformance Pack — 133-fixture black-box test suite; signed
kye.conformance_report.v1envelope. - Audit Chain — per-tenant WORM-anchored audit chain; the specific multi-tier immutability construction is part of the patent track and is not disclosed here.
- Compliance Attestation — per-framework signed
kye.compliance.attestation.v1envelopes (90-day cadence).
Adjacent paths
- All frameworks — the framework catalogue (this is a deep-dive).
- For regulators — what supervisors see.
- For auditors · Onboard your firm
- Whitepaper — the technical foundation.
- Apply for a regulated-pilot — banking-grade scoped engagement.