/ai-act-readiness

name: "ai-act-readiness"
description: "EU AI Act readiness: risk tier, duties, deadlines. Load whenever AI ships in the EU or the user asks if the Act applies."

/compliance-os:ai-act-readiness — EU AI Act Forcing Questions

**Command:** `/compliance-os:ai-act-readiness <system>`

The EU AI Act compliance operator pressure-tests any AI system before EU deployment. Six Article-cited questions before any EU placement, conformity assessment, or annual compliance refresh.

When to Run

• During AI-system intake review (per new system or material change)
• Before placing an AI system on the EU market
• Before signing the EU declaration of conformity (Article 47)
• During annual compliance refresh (Article 113 phasing brings new obligations)
• When the organization's role changes (deployer becomes provider via Article 25(1) substantial modification)
• When training compute approaches 10^25 FLOPs (Article 51 systemic-risk threshold)

The Six EU AI Act Questions

1. Article 5: Is this a prohibited AI practice?

**Penalty: up to 35M EUR or 7% worldwide turnover.**

• 8 categories: subliminal manipulation, exploitation of vulnerabilities, social scoring, predictive policing, untargeted facial scraping, emotion recognition in workplace/education, biometric categorisation by sensitive attributes, real-time public biometric ID by law enforcement
• Run `ai_system_risk_classifier.py`
• If yes → STOP. Cannot place on EU market. No exceptions outside Article 5(2) carve-outs.

2. Article 6 + Annex III: Is this high-risk?

**Annex III triggers high-risk; Article 6(3) carve-out conditional.**

• 8 categories: biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, justice
• Carve-out applies only if Article 6(3)(a)-(d) AND no profiling of natural persons
• Profiling overrides carve-out (Article 6(3) last sentence)
• Run `ai_system_risk_classifier.py`

3. Article 43: For high-risk, Module A or Module H?

**Annex III points 2-8 → Module A (internal control, Article 43(2)). Biometrics → provider's choice of Module A or H if harmonised standards or common specifications are applied in full, otherwise Module H (Article 43(1)).**

• Run `conformity_assessment_planner.py`
• Module A (Annex VI): internal control with presumption of conformity if Article 40 harmonised standards applied
• Module H (Annex VII): full QMS + notified body; optional for biometrics with standards applied, mandatory for biometrics where standards or common specifications do not exist, were not applied or only partly applied, or a standard was published with a restriction
• Annex IV technical documentation: 8 items required before placing on market

4. Article 25: What role does the company play?

**Provider obligations are heaviest; substantial modification turns deployer into provider.**

• Provider (Article 3(3)): placed on market; full Title III + Article 73 reporting
• Deployer (Article 3(4)): Article 26 obligations + Article 27 FRIA if public sector
• Importer (Article 3(6)): Article 23 verification of conformity
• Distributor (Article 3(7)): Article 24 CE marking verification
• Authorized representative (Article 22): non-EU providers must appoint
• Run `ai_act_obligation_tracker.py`

5. Article 50: Are transparency obligations satisfied?

**In force 2 Aug 2025.**

• Article 50(1): disclose AI interaction to natural persons (chatbots, virtual agents)
• Article 50(2): mark synthetic content as AI-generated
• Article 50(3): disclose emotion recognition / biometric categorisation (outside Article 5 prohibitions)
• Article 50(4): disclose deepfakes (image, audio, video) as AI-generated

6. Articles 51-55: Is this a GPAI? Does it have systemic risk?

**GPAI has parallel track; systemic risk above 10^25 FLOPs.**

• Article 3(63): general-purpose AI model definition
• Article 51: systemic-risk presumption (≥ 10^25 FLOPs training compute) or Commission designation
• Article 53: all GPAI providers — Annex XI technical docs, Annex XII downstream info, copyright policy, training-data summary
• Article 55: systemic-risk GPAI additional obligations — model evaluations, adversarial testing, incident reporting, cybersecurity
• Article 54: non-EU GPAI providers must appoint authorized representative

Workflow

# 1. Risk classification
python ${CLAUDE_PLUGIN_ROOT}/vendor/ra-qm-team/eu-ai-act-specialist/scripts/ai_system_risk_classifier.py systems.json

# 2. If high-risk: conformity assessment
python ${CLAUDE_PLUGIN_ROOT}/vendor/ra-qm-team/eu-ai-act-specialist/scripts/conformity_assessment_planner.py system.json

# 3. Per-role obligation matrix
python ${CLAUDE_PLUGIN_ROOT}/vendor/ra-qm-team/eu-ai-act-specialist/scripts/ai_act_obligation_tracker.py roles.json

# 4. Cross-framework reuse (ISO 42001 etc.)
python ../../skills/compliance-os/scripts/cross_framework_mapper.py program.json

Output Format

# EU AI Act Readiness: <system>
**Date:** YYYY-MM-DD
**Article Citations:** Every verdict below cites the specific Article.

## The Decision Being Made
[classify | conformity-route | obligation-scope | annual-refresh]

## Risk Classification
- Tier: prohibited | high_risk | limited_risk | minimal_risk
- Citation: Article X(Y) + Annex Z if applicable
- Rationale: <Article-cited rationale>
- GPAI: yes/no
- Systemic-risk GPAI: yes/no (per Article 51 10^25 FLOPs threshold)

## Conformity Assessment (if high-risk)
- Module: A | A_with_caveats | A_or_H | H | sectoral
- Citation: Article 43 + Annex VI/VII
- Notified body required: yes | no | provider_choice | depends_on_sectoral_law
- Annex IV pack status: complete | in-progress | not-started

## Obligation Matrix
- Total obligations: N
- By deadline phase: 2025-02-02=A, 2025-08-02=B, 2026-08-02=C (Art. 50), 2027-12-02=D (Annex III high-risk), 2028-08-02=E (Annex I high-risk); dates per Regulation (EU) 2026/1744
- Highest-priority unmet obligation: <Article + description>

## Transparency (Article 50)
- 50(1)