// compliance · Israel INDA + Ministry of Justice

MoorAI & Israel's
Responsible-AI Guide.

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.

Covered MoorAI provides the control directly Supports MoorAI supplies enforcement / evidence; the org owns the rest Org-led an organizational duty MoorAI assists but can't replace

Governance & roles

Guide elementWhat it asksHow MoorAI addresses itStatus
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.

Four-stage risk process

Guide elementWhat it asksHow MoorAI addresses itStatus
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.

Risk-mitigation toolset

Guide elementWhat it asksHow MoorAI addresses itStatus
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.

Product & organizational measures

Guide elementWhat it asksHow MoorAI addresses itStatus
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.

Inventory & reporting

Guide elementWhat it asksHow MoorAI addresses itStatus
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.

glick.run — AGPL-3.0