5.1 EU AI Act Technical File Template

System Name: Braintrust AIR – Automated Interview & Evaluation Platform
Use Case: Candidate screening and evaluation for employment purposes (Annex III, §4 – High-Risk AI)
Client: [Insert Client Name]
Prepared By: [Client IT Admin or Compliance Lead]
Date Created: [Insert Date]
Last Updated: [Insert Date]
Version: 1.0

1. System Overview

• System Description:
  Braintrust AIR conducts structured, asynchronous video interviews using AI-generated prompts. Candidate responses are evaluated through speech-to-text transcription against predefined job-specific scoring criteria.
   
• Deployment Scope:
   
  • Roles: [e.g., Software Engineer, Customer Support Agent]
  • Jurisdiction: [EU Countries of deployment]
  • Language Support: English, Spanish, German, Dutch, French, Japanese, Italian, Hindi, Portuguese, Russian
• Data Types Processed:
   
  • Candidate first/last name (for identification)
  • Interview audio and video (used for transcription)
  • Speech-to-text transcript (used for scoring)
  • AI-generated scorecards and rankings

2. Risk Management System (Art. 9)

• Risk Assessment Completed: ✅ Yes
• Risk Register Location: [Link to internal document or SharePoint]
• Risk Mitigation Measures:
  • Biannual Bias & Drift Audit (see §8 of AIR Compliance Doc)
  • Incident Response Plan (AIR Doc §7)
  • Human-in-the-loop review required for all candidate progressions

3. Data Governance & Fairness (Art. 10)

• Bias Mitigation Controls:
  • Structured scoring criteria defined by client
  • No use of biometric data (visual/audio features excluded from scoring)
  • Demographic-neutral evaluation via transcript-only analysis
• Bias Threshold: Δ ≤ 5% across demographic groups
• Accessibility: Client-managed accommodations for candidates with disabilities
• Audit Reports

4. Technical Documentation (Art. 11)

• Model Description & Version:
  • [e.g., AIR v1.8.2, LLM-based scoring engine tuned on domain-specific rubric]
• Training Data Summary:
  • Curated interview transcript corpus, anonymized, role-specific
  • No personal data from EU subjects used in model training

5. Record-Keeping (Art. 12)

• Log Retention:
  • Audit logs (interview timestamps, scoring decisions) retained ≥ 6 years
  • Stored in AWS
• Storage Location:
  • AWS - USA West Coast
• Access Control:
  • Role-based access managed via [SSO/MFA provider]

6. Transparency Measures (Art. 13)

• Candidate Notification:
  • Accessibility and AI usage disclaimer included in interview invite (Appendix A)
  • Link to FAQ
• Recruiter Training:
  • Mandatory AIR ethics and usage module (Module 1)
  • [Attach training deck or LMS link]

7. Human Oversight Protocol (Art. 14)

• Oversight Controls:
  • AI cannot auto-advance candidates
  • Human must review AIR scorecard and video before shortlisting

8. Accuracy, Robustness, and Cybersecurity (Art. 15)

• Performance Validation:
  • Quarterly Alignment Test comparing model outputs with recruiter evaluations
• Security Controls:
  • AES-256 encryption (at rest); TLS 1.2+ (in transit)
  • SOC 2 compliance (Braintrust); audit IN PROGRES
  • Penetration test reports [attached or linked]

9. Conformity Assessment Declaration (Art. 43)

• Assessment Type: Internal conformity assessment
• Status: ✅ Completed
• Signed Declaration:
  • [Insert scanned PDF or link]
  • Signed by: [Compliance Officer or Legal Representative]
  • Date: [Insert Date]

10. Registration in EU AI Database (Art. 51)

• Status: [☐ Pending / ✅ Completed]
• Upload Date: [Insert Date]
• Responsible Party: [Client IT Admin Name]
• Submission Location: [Link to EU AI Database or export copy]
  
   

Appendix A – Candidate Invite Disclaimer Template

How to Add Disclaimer

Recommended Inviter Emails

Appendix B – Audit Logs

[Link or attach bias audit, scoring review samples, SIEM extracts]

Appendix C – Training Material

[Attach recruiter training slides, usage SOP, and AI governance summary]