GreenLight for Thales

Independent AI Decision Assurance for Mission-Critical Workflows

An independent control layer that validates the evidence behind AI recommendations, evaluates applicable rules and delegated authority, and creates a reconstructable decision record before an action proceeds.

Select a demonstration scenario

First scenario

Authorization-Aware RAG

AUTH-RAG · synthetic corpus

The same question, asked as three users, retrieves different chunks. GreenLight enforces access when chunks are selected for the model.

Open authorization-aware RAG scenario

Primary demonstration

Predictive Maintenance Decision Assurance

MNT-2026-0042 · RADAR-042

An AI recommends deferring a radar inspection by 72 hours. GreenLight checks the technical baseline it relied on, the maintenance rules and who may approve the deviation.

Open maintenance scenario

cortAIx · C4I

Command & Control Decision Assurance

C2-2026-0017 · Sector 7

An AI decision aid recommends raising the monitoring priority of a surveillance sector. GreenLight verifies the source picture, operational rules, scope and release authority.

Open C2 scenario

Domain independence

Commercial Credit Exception

CR-2026-1088 · Maple Industrial Ltd.

Same interaction as the other two: outdated evidence, insufficient authority, and an authorized approval. A CAD 1.5M increase against a CAD 1M director limit.

Open credit scenario

Same Control Pattern, Different Domain

	Commercial credit	Predictive maintenance	C2 assurance
Decision	Credit exception	Maintenance deviation	Priority change
Subject	Borrower	Equipment / asset	Surveillance sector
Evidence	Financial statements	Telemetry · technical baseline	Fused picture · tasking
Rules	Credit policy	Engineering / maintenance rules	Operational governance rules
Requester	Director Credit	Maintenance Engineer	Watch Officer
Approver	VP Credit	Authorized Engineering Approver	Duty Operations Manager
Record	Credit decision record	Operational decision record	Operational decision record

The differentiation

GreenLight does not ask another AI model whether the original AI was correct.

It independently retrieves the underlying evidence, evaluates deterministic controls and verifies who is authorized to approve the proposed action.

Complements cortAIx — an independent control extension, not another AI platform.

Synthetic data and simulated connectors. Live integration, identity validation, source-system permissions and execution enforcement belong to a subsequent PoC.

Build GL-THALES-2026.09.30-1