Age Verification Performance & Compliance
Technical performance metrics meeting international regulatory standards for child protection
1. Introduction & International Compliance Overview
Global Age Assurance Provider: Agemin provides age verification technology to help regulated service providers comply with international child protection requirements across multiple jurisdictions.
We publish comprehensive performance metrics to enable our clients to meet their regulatory obligations under various international frameworks including record-keeping, transparency, and public disclosure requirements.
International Regulatory Compliance
| Jurisdiction | Regulatory Body | Framework | Key Requirements Met |
|---|---|---|---|
| 🇬🇧 United Kingdom | Ofcom | Online Safety Act 2023 | Age assurance for Part 5 services, transparency reporting |
| 🇩🇪 Germany | KJM | JMStV / JuSchG | Youth protection, closed user groups, age verification systems |
| 🇫🇷 France | ARCOM | Loi SREN / LCEN | Double anonymous age verification, data minimization |
| 🇺🇸 United States | FTC | COPPA | Verifiable parental consent, children under 13 protection |
| 🇪🇺 European Union | European Commission | Digital Services Act (DSA) | Risk assessment, age-appropriate design, transparency |
| 🇦🇺 Australia | eSafety Commissioner | Online Safety Act 2021 | Age verification, restricted access systems |
This transparency report provides detailed performance metrics that meet or exceed requirements across all major international age verification regulations. Our technology is designed to be compliant by default, allowing platforms to operate globally while protecting minors effectively.
Privacy-First
Meeting GDPR, COPPA, and global privacy standards with minimal data collection
Accuracy Standards
Exceeding technical requirements across all jurisdictions with 96.3% accuracy
Accessibility
WCAG 2.1 AA compliant with alternative verification methods
2. Description of Age Assurance Methods
Primary Method: Facial Age Estimation
Our system uses advanced computer vision to estimate a user's age from a selfie photograph. The process involves capturing a live image, analyzing facial features, and determining whether the person appears to be above or below the required age threshold.
Challenge-Age Approach (Optional): Platforms can enable a challenge threshold where users estimated below a certain age (e.g., 23 years for 18+ content) are required to complete additional verification through government-issued ID. This optional feature helps minimize false negatives and ensures higher accuracy for borderline cases.
Email Age Estimation - Privacy-First Method
Email address age assurance is recognized by Ofcom as a highly effective method that is private, compliant, and simple. The process analyzes a user's digital footprint, comparing the submitted email address against online databases and metadata sources to estimate their age.
For instance, an email address registered with a mortgage broker, roofing service, or professional networking sites could indicate an adult user. Historical data points such as account creation dates, service registrations, and public records provide age signals without requiring new personal information.
Key Privacy Advantage: Unlike facial age estimation, this method relies exclusively on data that users have already shared publicly or semi-publicly, eliminating the need to provide new personally identifying information (PII). No biometric data is collected, stored, or processed.
Liveness Detection & Anti-Spoofing
For facial verification methods, every check includes passive liveness detection to ensure the image is from a real person present at the time of verification. This prevents the use of photos, videos, masks, or other presentation attacks.
Fallback Method: Document Verification
When primary methods cannot provide a confident result, users can verify their age using government-issued identification. This includes passport, driver's license, or national ID card verification with authenticity checks.
3. Performance Metrics (Technical Accuracy)
Dataset Information: Metrics based on testing with 2.5 million diverse facial images across all demographics and ethnicities globally. Benchmark: IMDB-WIKI, FG-NET, and proprietary international datasets.
Testing Period: August - September 2025 | Sample Size: n=2,500,000
Geographic Coverage: Testing included populations from North America, Europe, Asia-Pacific, Latin America, Middle East, and Africa
| Metric | Value | Age Group | Description |
|---|---|---|---|
| MAE (Mean Absolute Error) | 1.1 years | 18-21 years | Average age prediction error for young adults |
| MAE (Mean Absolute Error) | 0.9 years | 9-17 years | Average age prediction error for minors |
| MAPE (Mean Absolute Percentage Error) | 7.8% | 13-17 years | Relative error rate for teens |
| TPR (True Positive Rate/Recall) | 98.2% | Under 18 | Percentage of under-18s correctly identified as minors |
| FPR (False Positive Rate) | 1.9% | 18+ | Percentage of adults incorrectly classified as minors |
| FNR (False Negative Rate) | 1.8% | Under 18 | Percentage of minors incorrectly classified as adults |
| SD (Standard Deviation) | 1.5 years | All ages | Prediction variability across all age groups |
| Overall Accuracy | 96.3% | 18 threshold | Correct classification for under/over 18 decisions |
High Accuracy
96.3% overall accuracy for 18+ threshold determination
Low Error Rate
1.8% false negative rate minimizes risk of minors accessing restricted content
User-Friendly
1.9% false positive rate ensures minimal friction for adult users
4. Robustness & Reliability
Lighting Conditions
Tested: Low light, bright light, backlight, artificial light
Camera Quality
Tested: 240p to 4K resolution, various devices
Face Angles
Tested: ±45° yaw, ±30° pitch, ±15° roll
Partial Occlusion
Tested: Masks, sunglasses, hair coverage
Presentation Attacks
Tested: Photos, videos, masks, deepfakes
Monitoring & Model Drift Prevention
- •Continuous monitoring of model performance with automated alerts for accuracy degradation
- •Monthly retraining cycles with new data to maintain performance standards
- •A/B testing of model updates before production deployment
- •Reproducibility testing ensures consistent results across deployments
5. Fairness & Bias
Performance Parity Achieved: Testing across all demographic groups shows performance variance within 2.3%, with no statistically significant bias detected. All groups receive equitable treatment in age verification decisions.
| Demographic Category | Groups Tested | Performance Variance | Result |
|---|---|---|---|
| Sex | Male, Female | ±1.8% | No significant bias |
| Ethnicity | White, Black, Asian, Hispanic, Other | ±2.1% | Performance parity achieved |
| Age Brackets | 9-12, 13-15, 16-17, 18-21, 22-25 | ±1.5% | Consistent across groups |
| Facial Features | Glasses, No glasses, Facial hair, Head coverings | ±2.3% | Within acceptable range |
Continuous Fairness Monitoring
We conduct quarterly fairness audits and publish updates to these metrics. Any demographic group showing performance variance exceeding 3% triggers immediate model retraining and bias mitigation procedures.
6. Privacy & Data Protection
Core Privacy Principle
We do not store biometric images or predicted ages. Only anonymized metadata (pass/fail result, verification method used, timestamp) is retained for audit purposes.
What We Store
- Pass/fail verification result
- Verification method used
- Timestamp of verification
- Session token (expires after use)
What We Don't Store
- ✕Facial images or biometric data
- ✕Actual or estimated ages
- ✕Personal identification details
- ✕Cross-site tracking data
Privacy Commitments
Global Privacy & Compliance Certifications
Our privacy and security practices comply with international standards:
Data Protection Impact Assessments (DPIA) conducted: November 2024 | Privacy by Design certified
7. Accessibility & Usability
WCAG 2.1 Level AA Compliance
CertifiedFull compliance with Web Content Accessibility Guidelines
Screen Reader Support
ImplementedCompatible with JAWS, NVDA, VoiceOver
Keyboard Navigation
Full SupportAll functions accessible via keyboard
Alternative Verification
AvailableID verification fallback for facial estimation failures
Multi-language Support
42 LanguagesInterface available in major languages
High Contrast Mode
SupportedEnhanced visibility for visual impairments
Alternative Verification Routes
When facial age estimation cannot be completed due to accessibility needs or technical limitations, users can access alternative verification methods:
- •Government ID verification with OCR text extraction
- •Credit card verification (where legally permitted)
- •Mobile network operator age verification
- •Manual review process for edge cases
8. Updates & Transparency
Update Schedule
Testing Commitments
- Annual comprehensive re-testing of all metrics
- Quarterly fairness and bias assessments
- Immediate re-testing after major model updates
- Public disclosure of material changes
Further Information
For regulated service providers requiring additional compliance documentation or detailed technical specifications, please contact our compliance team.
This transparency report contains performance metrics that meet or exceed international regulatory standards including Ofcom (UK), KJM (Germany), ARCOM (France), COPPA (USA), and DSA (EU) requirements. Agemin maintains compliance through continuous testing and third-party audits rather than specific regulatory certifications.