Sentient Foundation has launched Fingerprinting Leaderboard, an open benchmark, along with the open-source toolkit mlprints, to evaluate AI model fingerprinting techniques and their resilience to adversarial attacks. According to Odaily, the benchmark is designed to test how well model fingerprints can be embedded and removed, and to help developers preserve proof of ownership while releasing open model weights.

Sentient Labs said research presented at SaTML 2026 showed that 10 existing model fingerprinting techniques, including its Perinucleus method, could be bypassed under complex adaptive attacks. The benchmark is intended to provide open testing for fingerprinting effectiveness and resistance to removal.