Skip to content

EU AI Act Compliance

How a2p supports EU AI Act requirements.

Overview

The EU AI Act establishes rules for AI systems. a2p helps agents comply through:

  • Risk classification in agent profiles
  • Transparency obligations built-in
  • Human oversight mechanisms
  • Technical documentation support

Risk Classification

Agent profiles include AI Act risk level:

{
  "aiActCompliance": {
    "riskLevel": "limited",
    "transparencyObligations": [
      "user_notification",
      "ai_generated_content_marking"
    ]
  }
}

Risk Levels

Level Examples Requirements
Minimal Spam filters None
Limited Chatbots Transparency
High CV screening Full compliance
Unacceptable Social scoring Prohibited

Transparency (Article 52)

User Notification

Agents must disclose AI nature:

{
  "transparency": {
    "isAI": true,
    "disclosure": "I am an AI assistant powered by..."
  }
}

Content Marking

AI-generated content must be marked:

{
  "transparencyObligations": [
    "ai_generated_content_marking"
  ],
  "contentMarking": {
    "method": "watermark",
    "visible": true
  }
}

Human Oversight

For high-risk systems:

{
  "humanOversight": {
    "required": true,
    "mechanism": "human_on_the_loop",
    "escalationThreshold": 0.7,
    "reviewProcess": "Manual review of flagged decisions"
  }
}

Oversight Mechanisms

Type Description
human_in_the_loop Human approves each decision
human_on_the_loop Human monitors, can intervene
human_in_command Human can override anytime

Conformity Assessment

For high-risk systems:

{
  "conformityAssessment": {
    "completed": true,
    "assessor": "Notified Body XYZ",
    "date": "2025-06-15",
    "certificateId": "CE-2025-12345",
    "validUntil": "2030-06-15"
  }
}

Technical Documentation

a2p profiles can link to required documentation:

{
  "documentation": {
    "technicalDocumentation": "https://example.com/docs/tech",
    "riskAssessment": "https://example.com/docs/risk",
    "testingResults": "https://example.com/docs/tests"
  }
}

Implementation Checklist

  • Risk level declared in agent profile
  • Transparency obligations listed
  • AI disclosure implemented
  • Content marking (if applicable)
  • Human oversight (if high-risk)
  • Conformity assessment (if high-risk)
  • Documentation links provided

Next Steps