Sentient launches open benchmark to test AI model fingerprinting

Sentient Foundation has launched the Fingerprinting Leaderboard, an open benchmark for evaluating AI model fingerprinting techniques. The move follows Sentient Labs research showing that nearly all ten tested schemes, including its own Perinucleus method, could be evaded by adaptive attacks. An open-source library, mlprints, ships alongside the leaderboard.

• Benchmark measures fingerprint embedding without hurting model quality
• Adversarial robustness testing is planned for a later release
• Perinucleus embedded 24,576 fingerprints into Llama-3.1-8B
• mlprints library unifies methods and attacks in one environment