EU AI Act Compliance¶
How a2p supports EU AI Act requirements.
Agent Classification¶
Agent profiles include AI Act fields:
{
"aiActCompliance": {
"riskLevel": "limited",
"transparencyObligations": [
"user_notification",
"ai_generated_content_marking"
],
"humanOversight": {
"required": true,
"mechanism": "human_on_the_loop"
}
}
}
Risk Levels¶
| Level | Description | a2p Support |
|---|---|---|
| Minimal | Low risk AI | ✅ Basic profile |
| Limited | Transparency needed | ✅ Transparency fields |
| High | Significant impact | ✅ Full compliance fields |
| Unacceptable | Prohibited | ❌ Not supported |
Transparency (Article 52)¶
| Obligation | Implementation |
|---|---|
| User notification | transparencyObligations array |
| AI content marking | Declared in profile |
| Capability disclosure | Agent capabilities field |
Human Oversight (Article 14)¶
{
"humanOversight": {
"required": true,
"mechanism": "human_on_the_loop",
"interventionCapability": true
}
}
Memory proposals implement human-in-the-loop by default.
Next Steps¶
- GDPR — GDPR compliance
- DPIA Guide — Assessment guide