Collected molecules will appear here. Add from search or explore.
Provides a framework for verifying that specific data has been removed from a machine learning model (Machine Unlearning) using Membership Inference Attacks (MIA) as oracles and recording the proof on a blockchain for GDPR compliance auditing.
stars
0
forks
0
VeriForgot is a research-oriented project aimed at the intersection of privacy law (GDPR) and machine learning. Quantitatively, the project has zero traction (0 stars, 0 forks), indicating it is likely a supplemental repository for a paper submission (CRBL 2026). Its defensibility is currently minimal; it functions as a proof-of-concept rather than a production-ready tool. The core 'moat' would be the specific algorithmic approach to using Membership Inference Attacks (MIA) as a verification oracle, but MIA techniques are well-documented in academia and easily replicated. Frontier labs are unlikely to adopt blockchain for model auditing, as they prefer centralized compliance frameworks or differential privacy. The primary risk is academic displacement; machine unlearning is a fast-moving field, and MIA-based verification is often criticized for providing statistical rather than absolute guarantees of data removal. Competitors include established unlearning frameworks like SISA (Sharded, Isolated, Sliced, and Aggregated) and emerging privacy-preserving ML libraries from OpenMined.
TECH STACK
INTEGRATION
reference_implementation
READINESS