00Article 50 Transparency Readiness

Compliance is not a checkbox.
It is an evidence trail.

Map obligations to the evidence you already have. Reveal what is supported, missing, conflicting or still requires human interpretation.

Evidence-readiness support, not legal advice or certification.
Built during OpenAI Build Week

No setup required. All demo inputs are editable.

01Scope Studio

Decide what is worth reviewing — not what is compliant.

Each answer changes which controls may require review. It does not determine that any regulation applies.

Does the system interact directly with natural persons?

Governs whether Article 50(1) interaction disclosure and Article 50(5) clarity controls surface.

Does it generate or manipulate text, audio, image or video?

Governs whether Article 50(2) machine-readable marking surfaces.

Is the organisation acting as provider, deployer or both?

Provider, deployer or both. Determines which obligations attach.

Is the output used in the European Union?

Territorial scope of the Regulation.

Are exceptions or special contexts under specialist review?

Law-enforcement, safety and other carve-outs need qualified interpretation.

02Source Room

Three descriptions of one system.

Do they tell the same story?

System

AI System Description

5 lines
  1. 1LuminaDesk Assistant is a generative AI workplace assistant available through a conversational interface.
  2. 2It answers employee questions and generates draft text.
  3. 3The product team describes the system as “an intelligent workplace expert.”
  4. 4The system may produce inaccurate content and users are advised to verify important information.
  5. 5Human review is available for selected enterprise workflows.
Original
Technical

Technical & Product Documentation

6 lines
  1. 1The interface sends user prompts to a generative model and returns generated text.
  2. 2The application currently displays the label “LuminaDesk Assistant.”
  3. 3The product specification requires an AI interaction notice during onboarding.
  4. 4The current implementation checklist does not contain verification evidence for that onboarding notice.
  5. 5Generated text is stored with internal provenance metadata.
  6. 6The documentation does not demonstrate an interoperable machine-readable output marking mechanism.
Original
User-facing

User-Facing Disclosure

4 lines
  1. 1Meet LuminaDesk, your workplace expert.
  2. 2Ask questions and receive immediate answers based on your organisation’s knowledge.
  3. 3The interface does not explicitly state in its welcome screen that the user is interacting with an AI system.
  4. 4Public documentation states that “all answers are verified,” while internal documentation advises users to verify important information.
Original
Run evidence mapping

A deterministic local pass. No network requests, no runtime AI, no data leaves your browser. Line references are exact.

Sources & scope

Primary references remain authoritative.

According to current European Commission materials, the selected Article 50 transparency obligations apply from 2 August 2026. This demo covers selected transparency-readiness controls only. Official law and guidance remain authoritative.

Provenance

Built during OpenAI Build Week

Prior foundation

AI Review Engine previously explored:

  • requirement matching
  • documentation conflicts
  • unsupported claims
  • evidence trails
  • verification leads rather than verdicts

New Build Week contribution

  • standalone AI Compliance Evidence Mapper
  • selected Article 50 readiness control model
  • applicability questionnaire
  • regulatory and internal-control distinction
  • evidence-strength classification
  • evidence-to-control mapping
  • conflict comparison
  • ownership routing
  • deterministic action planning
  • evidence coverage model
  • two organisational evidence states
GPT-5.6

Product and reasoning architecture

Used to identify the narrow evidence-readiness MVP, distinguish readiness from legal compliance, design the control–evidence–gap–owner–action architecture, create the fictional scenario and establish product and safety boundaries.

The demo does not make runtime GPT-5.6 calls.

Codex

Engineering acceleration

Used to scaffold and implement the standalone application, build the deterministic evidence engine, create the versioned control model, implement references and conflict detection, write automated tests, improve accessibility and responsiveness, and verify the production build.

Final product, architecture and legal-boundary decisions remained human-led.