Israel's National Digital Agency and Ministry of Justice published a "Guide for Responsible Use of AI in the Public Sector" (v1.0, 2026) — a governance and risk-management framework built on the OECD trustworthy-AI principles. It defines governance roles, a four-stage risk process, and a toolset for mitigating each AI risk. MoorAI is the on-device, content-free enforcement + evidence layer that operationalizes much of that toolset. Below is an honest map of what MoorAI covers, what it supports, and what stays org-led.
| Guide element | What it asks | How MoorAI addresses it | Status |
|---|---|---|---|
| Organizational AI lead + governance forum | Appoint an AI lead and a cross-functional forum to own responsible-AI strategy. | Org-led role. MoorAI supplies the live inventory, risk signals and evidence the lead and forum govern with. | Org-led |
| Business-application owner | Each use case's owner runs its risk management and monitoring. | Per-agent assurance score, enforcement and signed records give the owner live control + proof. | Supports |
| End-user responsible-use guidelines | Staff must not paste restricted data, must keep oversight, verify, and report. | On-device content-free DLP + coach/alert enforce the do's/don'ts at the point of use. | Covered |
Source: Section 3, Appendices A–B.
| Guide element | What it asks | How MoorAI addresses it | Status |
|---|---|---|---|
| Stage 1 — analyze & triage | Rate each AI use case's risk and route it (green/yellow/red/blue). | Discovery + AIBOM reveal what's actually running to triage; per-device policy tiers. | Supports |
| Stage 2 — build a mitigation plan | Select controls proportionate to the risk. | MoorAI is the enforcement toolbox the plan draws on (see the toolset below). | Supports |
| Stage 3 — operate with monitoring & sample QA | Run with enforcement, monitoring, and periodic output checks. | Coach/alert/block enforcement, output review, assurance score, content-free monitoring. | Covered |
| Stage 4 — AI incident response | Detect, contain, roll back, learn. | Content-free signals (category·risk·hash), signed decisions and lineage feed the AIRT; break-glass to contain. | Supports |
Source: Section 4.
| Guide element | What it asks | How MoorAI addresses it | Status |
|---|---|---|---|
| Prompt injection & info-security | Guardrails, model-armor, leakage protection, on-prem models. | On-device pre-flight review of prompts + lethal-trifecta / toxic-flow detection; content-free, so no leakage. | Covered |
| Loss of control over agents | Govern MCP / A2A tool connections. | On-device MCP tool-call gateway, server allow-list, per-tool argument rules, JIT elevation. | Covered |
| Accuracy & model drift | Tool-use limits, model allow-list, RAG/CAG grounding, drift monitoring. | Model-endpoint allow-list + output review + assurance score. Grounding & fact-checking stay with the org. | Supports |
| Privacy & disclosure | Privacy-enhancing tech (PETs), anonymization, on-prem. | Content-free by construction — the on-device analog of PETs; only category·risk·hash ever leave. | Covered |
| Bias & fairness | Fairness testing, debiasing, bias-evaluation frameworks. | Human-oversight enforcement + content-free audit trail. MoorAI is not a bias-testing tool. | Org-led |
| Transparency & explainability | XAI, model cards / datasheets. | Signed decisions + Event-Flow lineage + AIBOM are the evidence side of transparency. | Supports |
Source: Appendix C.4 — technical & architectural.
| Guide element | What it asks | How MoorAI addresses it | Status |
|---|---|---|---|
| Disclosure to users | The system tells users it's AI-based and states its limits. | Coach / disclosure prompts surfaced at the point of use. | Supports |
| User feedback / reporting | A central mechanism to report problems and harms. | Justification & report signals feed the console — content-free. | Supports |
| Human oversight | Human in / on / after the loop for consequential use. | Coach / alert / block keeps a human in the loop; ex-ante and real-time enforcement. | Covered |
| Data governance | Data quality, access control, bias checks on the data feeding AI. | Content-free telemetry + model/endpoint allow-list keep sensitive data on the device. | Supports |
| Break-glass / blue-track | Time-boxed, reversible, disclosed trials for higher-risk use. | Break-glass / offline fail-closed + per-device policy tiers. | Covered |
Source: Appendix C.4 — product & business-organizational.
| Guide element | What it asks | How MoorAI addresses it | Status |
|---|---|---|---|
| Organizational AI system registry + status monitoring & rating | Maintain a live inventory of AI systems with a risk rating and status. | AIBOM + shadow-AI discovery + assurance score = a live, rated registry of what's running. | Covered |
| Internal reporting on all operating & shadow AI | Report on every AI system in use, including the unmanaged ones. | Shadow-AI discovery + content-free reporting surface the unmanaged installs. | Covered |
Source: Appendix A.3 — org-level operation.
Honest bottom line. MoorAI directly covers most of the guide's technical risk-mitigation toolset (prompt-injection defense, agent/MCP control, privacy via content-free design, human oversight, break-glass) and its inventory and end-user requirements. It supports the risk process and transparency with signed, content-free evidence. It does not replace the organizational governance the guide requires — appointing an AI lead, standing up a forum, writing policy, fairness testing and staff training — but it operationalizes and evidences them. This page is an independent analysis for planning; it is not legal advice or a certification, and references reflect the guide as published.