EU AI Act Compliance Checker
Analyze content against EU AI Act (Regulation 2024/1689) requirements.
Overview
The EU AI Act compliance checker validates content against:
- Article 5: Prohibited AI practices
- Article 6 / Annex III: High-risk system classification
- Article 14: Human oversight requirements
- Article 52: Transparency obligations
Installation
// Available in @sentinelseed/vscode extension
import { EUAIActChecker, checkEUAIActCompliance } from '@sentinelseed/vscode';
Quick Start
const checker = new EUAIActChecker();
// Basic heuristic check (100% local, no network)
const result = checker.check(content);
console.log(result.compliant); // true or false
console.log(result.riskLevel); // 'minimal', 'limited', 'high', 'unacceptable'
console.log(result.prohibitedPractices);
console.log(result.recommendations);
Article 5: Prohibited Practices
| Practice | Description |
|---|
| Social scoring | Evaluating trustworthiness based on social behavior |
| Biometric categorization | Classification by race, religion, sexuality |
| Emotion recognition | In workplace/education contexts |
| Predictive policing | Individual crime risk prediction |
| Subliminal manipulation | Below conscious awareness |
| Vulnerability exploitation | Targeting children, elderly, disabled |
| Real-time biometric ID | Mass surveillance in public spaces |
Annex III: High-Risk Systems
| Context | Examples |
|---|
| Biometrics | Facial recognition, fingerprint matching |
| Critical Infrastructure | Power grid, water supply, traffic |
| Education | Admission decisions, grade prediction |
| Employment | Recruitment, performance monitoring |
| Essential Services | Credit scoring, insurance, benefits |
| Law Enforcement | Criminal investigation, profiling |
| Migration | Asylum decisions, visa processing |
| Justice | Sentencing, legal outcome prediction |
Risk Levels
| Level | Meaning |
|---|
minimal | General AI systems |
limited | Transparency requirements |
high | Conformity assessment required |
unacceptable | Prohibited under Article 5 |
Human Oversight Models
| Model | Description |
|---|
human_in_the_loop | Human approves each decision |
human_on_the_loop | Human can intervene |
human_in_command | Human maintains control |
Compliance Result
interface EUAIActComplianceResult {
compliant: boolean;
riskLevel: 'minimal' | 'limited' | 'high' | 'unacceptable';
systemType: 'minimal_risk' | 'high_risk' | 'prohibited';
oversightRequired?: 'human_in_the_loop' | 'human_on_the_loop' | 'human_in_command';
prohibitedPractices: string[];
highRiskContexts: string[];
articleFindings: ArticleFinding[];
recommendations: string[];
}
References