FIT-Print: Towards False-claim-resistant Model Ownership Verification via Targeted Fingerprint
Researchers have developed FIT-Print, a novel method for verifying ownership of open-source AI models that is resistant to false claim attacks. Existing fingerprinting techniques are vulnerable to adversaries falsely claiming ownership of independent models. FIT-Print addresses this by using targeted signatures derived from model outputs and feature attributions, achieving a 100% defense success rate against false claims and a 0.0% false alarm rate on independent models. AI
IMPACT Enhances security for open-source AI models by preventing fraudulent ownership claims.