Frequently Asked Questions
Answers to common questions about AI Verification Service
General Questions
What is the AI Verification Service?
The AI Verification Service is an automated copyright verification platform that uses multiple AI providers to analyze submitted musical works and recordings. It detects copyright infringement, laundering attempts, AI-generated content, and duplicate registrations. The service is designed to assist administrators in making informed decisions during the copyright registration review process.
Who can use the AI Verification Service?
The service is restricted to administrators only. Regular users cannot access the verification functionality. You must have a valid SempreID with administrator privileges to use the service. Access is granted through the CopyrightChains Dashboard after administrator status has been verified.
How much does it cost to use?
The AI Verification Service is currently provided at no additional cost to administrators as part of the CopyrightChains ecosystem. There are no per-verification fees or subscription charges at this time. Usage is limited by rate limiting (100 requests per hour) to ensure fair access for all administrators.
Is verification mandatory for all registrations?
While not technically mandatory at the system level, it is strongly recommended that all copyright registrations undergo AI verification before approval. This helps ensure the integrity of the CopyrightChains registry and protects against fraudulent or infringing content. Many administrators make verification a required step in their approval workflow.
Can I access the service outside the Dashboard?
Yes, the service provides a REST API for programmatic access. However, you still need valid SempreID authentication and administrator privileges to use the API. The API is primarily intended for custom integrations and automated workflows. See the API Integration documentation for details.
Verification Process
How long does a verification take?
Most verifications complete within 12-60 seconds depending on content complexity and provider availability. Simple verifications with minimal content may finish in 12-20 seconds. Complex analyses with multiple large files may take 45-90 seconds. If processing exceeds 2 minutes, check the troubleshooting guide for timeout issues.
What content is analyzed?
The service analyzes all submitted content including audio files, lyrics, score sheets, album art, and metadata. Audio undergoes waveform analysis, fingerprinting, and spectral analysis. Text content is checked for similarity with known works. Images are analyzed for copyright indicators. Metadata is validated for consistency and authenticity.
How accurate is the AI detection?
The service uses multiple AI providers to improve accuracy, but no AI system is perfect. Confidence scores indicate the strength of detection - higher scores mean greater certainty. The service is designed to assist human decision-making, not replace it. Always review detailed findings and use human judgment, especially for medium-risk results.
Can I run verification multiple times?
Yes, you can run verification multiple times on the same registration. This is useful if you've updated content, received additional files, or want to confirm results. Note that results may vary slightly between runs due to database updates, provider availability, or AI model improvements. Each verification creates a new report with a unique ID.
What happens to my submitted files?
Files are temporarily stored during verification processing (typically 12-90 seconds), then automatically deleted. The service does not maintain long-term storage of submitted content. Only verification reports and metadata are retained in the database. All data transmission uses HTTPS encryption. Provider API calls use anonymized data when possible.
Understanding Results
What do the risk scores mean?
Risk scores range from 0-100. Low risk (0-30) indicates legitimate content with minimal concerns. Medium risk (31-70) suggests some issues that warrant further investigation. High risk (71-100) indicates significant concerns recommending rejection. The overall risk score combines all four verification types into a single metric.
Should I always follow AI recommendations?
No. AI recommendations provide guidance, but final decisions rest with administrators. Use AI as a powerful tool to inform decisions, but apply human judgment considering factors the AI cannot assess. Document your reasoning when making decisions that contradict AI recommendations. Report false positives and negatives to help improve the system.
What if different providers disagree?
Provider disagreement is normal since each analyzes different aspects. Review detailed findings to understand why providers reached different conclusions. Look for consensus among the majority. Consider which provider specializes in the relevant analysis type. When providers strongly disagree, conduct additional investigation before making final decision.
How do I interpret authenticity scores?
Authenticity scores range from 0-100, where higher is better. Scores above 70 indicate high confidence in human-created content. Scores 40-69 suggest uncertainty or mixed signals. Scores below 40 indicate likely AI-generated or heavily manipulated content. Authenticity below 40 typically warrants rejection regardless of other scores.
What causes low confidence scores?
Low confidence can result from insufficient content for analysis, poor file quality, unusual music styles, missing metadata, or corrupted files. To improve confidence: provide complete submissions with high-quality audio (320kbps MP3 or better), include lyrics and scores when available, ensure accurate metadata, and verify files aren't corrupted.
Technical Questions
Which AI providers are used?
The service integrates multiple AI providers: OpenAI (GPT-4o for text, Whisper for audio, Vision for images), Anthropic Claude (deep reasoning and analysis), Google Cloud AI (OCR, image labeling, speech-to-text), and custom ML models (audio fingerprinting, waveform analysis). Using multiple providers improves accuracy and provides redundancy.
What file formats are supported?
Audio: MP3, WAV, FLAC, AAC, OGG (max 100MB). Documents: PDF, TXT, DOCX (max 10MB). Images: PNG, JPG (max 5MB). For best results, use high-quality audio (320kbps MP3 or lossless), PDF for lyrics/scores, and clear images. Compressed formats like MP3 are preferred over large uncompressed formats.
How is my data protected?
All transmissions use HTTPS encryption. Files are stored encrypted during processing, then automatically deleted. No persistent storage of sensitive content. Provider API calls use anonymized data when possible. Admin-only access enforced through SempreID authentication. Audit logging tracks all verification requests for compliance.
What is the rate limit?
The service limits requests to 100 per hour per administrator to prevent abuse and ensure fair access. This includes all API calls (submit, status checks, report retrieval). If you exceed the limit, wait before making additional requests. For legitimate high-volume needs, contact support to discuss increased limits.
Can the service detect all types of infringement?
The service detects many infringement types but is not comprehensive. It excels at identifying similarity in melodies, lyrics, chord progressions, and waveform patterns. It may miss subtle infringement, complex arrangements, or works not in its reference database. The service is a tool to assist human review, not a complete replacement.
Specific Detection Types
How does AI-generated music detection work?
The service analyzes multiple indicators: vocal synthesis markers, artificial instrument patterns, lack of natural recording artifacts, algorithmic composition fingerprints, and metadata inconsistencies. It recognizes output from common AI music platforms (Suno, Udio, AIVA, Amper). Detection is most accurate with recent AI-generated content but may miss older or custom AI tools.
What is copyright laundering?
Copyright laundering is the practice of disguising existing copyrighted content as original work through minimal modifications. Common techniques include pitch shifting, tempo changes, simple remixing, re-recording with minor variations, and format conversion. The service detects these by analyzing pitch/tempo-invariant signatures, structural fingerprints, and metadata manipulation indicators.
How accurate is duplicate detection?
Exact duplicate detection is very accurate (near 100%) using cryptographic hashing. Near-duplicate detection is less precise, with accuracy depending on similarity level. The service catches obvious duplicates and flagging suspicious near-duplicates for review. False positives can occur with similar-sounding but distinct works. Always review duplicate findings before rejecting.
Can it detect unauthorized covers or remixes?
Partially. The service can identify when a recording is based on an existing musical work (cover) or heavily samples existing content (remix). However, it cannot determine whether proper licensing or permissions have been obtained. Covers and remixes may show high infringement risk - administrators must verify the applicant has proper authorization before proceeding.
Does it check against public databases?
The service primarily checks against the CopyrightChains internal database of registered works. It does not automatically cross-reference external databases like ASCAP, BMI, or copyright.gov. Administrators should conduct external database searches separately for medium-risk and high-risk cases requiring additional investigation.
Troubleshooting
What if verification times out?
Timeouts typically occur with very large files, provider outages, or queue overload. Check service health endpoint, wait 5 minutes and retry, compress large files, or try during off-peak hours. If timeouts persist, the issue may be with a specific AI provider - check provider status and accept partial results from successful providers.
Why is my report missing?
Try hard-refreshing the page (Ctrl+Shift+R), checking via API endpoint directly, re-authenticating to Dashboard, or clearing browser cache. If report truly missing after verification showed "completed", contact support with verification ID. Reports should always be generated for completed verifications.
What if I get authentication errors?
Ensure you're logged into Dashboard with valid SempreID session. Verify your account has administrator privileges. Enable cookies from copyrightchains.com domain. Try incognito mode to rule out browser extension interference. If problem persists, contact administrator to confirm your access level.
Can I cancel a running verification?
The service does not currently support cancellation of in-progress verifications. However, verifications typically complete quickly (under 60 seconds). If a verification appears stuck in processing state for more than 2 minutes, it may have failed. Check status endpoint or wait for timeout, then submit new verification.
How do I report false positives?
Document the false positive with verification ID, specific incorrect findings, and why you believe AI was wrong. Submit through Dashboard support feature or email [email protected]. Include supporting evidence demonstrating legitimate content. Your feedback helps improve AI accuracy for future verifications.
Best Practices
Should I verify every registration?
Yes, it's strongly recommended to verify all registrations before approval. Verification provides an important quality assurance checkpoint and helps maintain registry integrity. The small time investment (12-60 seconds processing) provides significant protection against fraud, infringement, and AI-generated content.
What should I do with medium-risk results?
Medium-risk results (31-70) require investigation. Review detailed findings to identify specific concerns. Research external sources for corroboration. Contact applicant for clarification or additional evidence. Consult senior administrators for complex cases. Document your investigation process and final decision reasoning.
How should I document verification decisions?
Record verification ID in registration notes. Summarize key findings from report. Document reasoning for approval/rejection. Note deviations from AI recommendations with explanation. Keep communication records with applicants. Track patterns for process improvement. Maintain audit trail for compliance.
Can I batch process verifications?
Yes, you can submit multiple verifications sequentially. The queue system handles them automatically with maximum 5 concurrent processing. Group similar registration types together. Process low-risk items first for efficiency. Flag medium/high-risk items for detailed review. Monitor queue metrics to avoid bottlenecks.
What makes a good submission for verification?
High-quality audio files (320kbps MP3 or better), complete lyrics and scores when available, accurate metadata (duration, format, technical details), clear album art, all files within size limits, no corrupted or damaged files. Better input quality produces more reliable verification results.
Future Development
Will new features be added?
Yes, the service is actively developed. Planned features include enhanced AI model training, expanded provider integration, real-time streaming analysis, batch verification API, improved reporting with visualizations, custom detection rules, webhook notifications, and public API access (with payment). Check release notes for updates.
Can I request specific features?
Yes, feature requests are welcome. Submit through Dashboard support or email with detailed description of desired functionality, use case explaining how it helps, priority level, and any relevant examples. Highly requested features are prioritized in development roadmap. Community feedback shapes service evolution.
Will the service support video content?
Video verification is planned for future development. This would include visual analysis of music videos, performance recordings, and audiovisual works. Implementation timeline depends on demand and technical requirements. Currently focused on audio and document verification as primary use cases.
Still Have Questions?
If your question wasn't answered here, explore these resources:
Getting Started
Introduction to the service
Verification Types
Detailed detection capabilities
Using the Service
Complete workflow guide
Troubleshooting
Common issues and solutions
Contact Support
For questions not covered in documentation:
- Dashboard Support: Use "Help" button in Dashboard
- Email: [email protected]
- Service Health: https://ai.copyrightchains.com/api/v1/health