ArticleGate

EU AI Act Compliance Gate for AI Agent Actions — deterministic article-level compliance verdict.

One-sentence pitch

ArticleGate answers: Does this AI agent action violate any Article of the EU AI Act?

Why this matters

The EU AI Act enforcement phase began August 2, 2026. High-risk AI obligations are law. But no product exists that deterministically maps an agent's action to specific articles and emits a compliance verdict at runtime. ArticleGate closes this gap.

Built from explore-pipeline insight: Regulatory Runtime Gap — governance frameworks exist but runtime enforcement doesn't.

What it is not

• Not legal advice or a substitute for compliance officers.
• Not an agent runtime or orchestration platform.
• Not a full AI governance suite.

It is a deterministic gate — check the article, emit the verdict, log the audit trail.

Articles modeled

Article	Title	What it checks
Art.5	Prohibited Practices	Subliminal manipulation, social scoring, real-time biometric ID
Art.6	High-Risk Classification	Context domains, action types, data categories
Art.13	Transparency	Transparency provided for high-risk / autonomous decisions
Art.14	Human Oversight	Oversight level (HITL/HOTL/HIC) for high-risk actions
Art.50	GPAI Transparency	AI-generated content labeling

Verdict scheme

Verdict	Meaning
compliant	All applicable articles passed
risk_flagged	Obligations triggered — mitigation required
prohibited	Article 5 violation — action blocked

Install / Run

Requires Python 3.10+ and no external packages.

python -m pip install -e .
python -m ArticleGate sample --out examples
python -m ArticleGate evaluate --action examples/compliant_action.json
python -m ArticleGate report --input verdict.json

Python API

from ArticleGate import evaluate_action, render_report

action = {
    "agent_id": "hr-screener-02",
    "action_type": "decision_recommendation",
    "risk_level": "high",
    "human_oversight": "none",
    "transparency_provided": False,
    "data_accessed": [{"category": "personal_sensitive"}],
    "deployment_context": "employment",
}

result = evaluate_action(action)
print(result["verdict"])  # "risk_flagged"
print(render_report(result))

Tests

python -m unittest discover -s tests -q

License

MIT License — see LICENSE.