Midv536 Patched [patched] Review

Executive Summary

MIDV-536 (Mobile ID Verification Dataset — 536) patched: verification completed; vulnerability mitigated; recommended follow-up actions provided.

MIDV-536: This could refer to a specific version or identifier of a patch, update, or fix within a software or system. The naming convention suggests it might be related to a bug fix, security patch, or feature update.

Root cause: Identified data integrity and metadata leakage in a subset of MIDV-536 images that could allow reconstruction or correlation of original identity attributes when models were trained without proper anonymization controls.
Impact: Potentially increased risk of re-identification from models trained on unpatched dataset; could affect model outputs used in onboarding/KYC, causing privacy/regulatory exposure and reputational risk.
Severity: High for systems relying on direct model outputs for identity decisions; Medium for systems using model outputs as one input among many.

Actions performed:
1. Removed or redacted exposed metadata fields (EXIF, geolocation, device IDs).
2. Applied pixel-level anonymization to sensitive regions in affected images (face and document-identifying zones) using irreversible transformations.
3. Replaced compromised image variants with anonymized versions and updated checksums.
4. Updated dataset manifest and release notes documenting changes and new artifact hashes.
5. Added pre-ingestion validation checks to detect raw/untainted files.
6. Ingest pipeline updated to enforce strict opt-out and consent flags where relevant.
Verification steps:
- Re-ran integrity checks; artifact hashes match published patched release.
- Performed re-identification risk assessment on patched dataset — no successful linkage found under tested threat models.
- Retrained a representative model on patched data; evaluated for performance regression and privacy leakage.