Launched this week
Fingerprinting Leaderboard
Open benchmark for robust, verifiable model fingerprints
5.0•21 reviews•49 followers
Open benchmark for robust, verifiable model fingerprints
5.0•21 reviews•49 followers
Models cost millions to train, but minutes to copy, erasing their origin traill. Introducing the Fingerprinting Leaderboard, an open benchmark for testing model provenance after copying, fine-tuning and distilling. The Sentient Foundation’s goal is to build the proof into the model itself.
Hello Product Hunt! 👋
At Sentient Foundation, we believe AI should belong to the people. Today we're bringing our Fingerprinting Leaderboard to a whole new community of builders.
Once a model's weights are out, no one can trace them. A model can be copied, fine-tuned or distilled, and its creator has no way to prove where it came from.
Our Sentient Labs researchers from Princeton, UW and UIUC spent two years on open-model IP protection. Their SaTML 2026 spotlight paper showed that a skilled attacker can evade almost every fingerprinting scheme. Their NeurIPS 2025 method, Perinucleus, embedded 24,576 fingerprints into Llama-3.1-8B with minimal performance loss.
The Fingerprinting Leaderboard is an open benchmark that compares how reliably fingerprinting methods verify provenance, survive attacks and preserve model utility. The first release ranks 14 schemes from Microsoft, UW, ETH Zurich, Tokyo Tech and China Agricultural University.
It's free and fully open source:
🔍 Explore and rank your scheme: https://fingerprint.sentient.fou...
📄 Methodology: https://github.com/sentient-agi/...
📚 Blog: https://sentient.foundation/news...
Tell us what our evaluation gets wrong, or which scheme you think is misranked!
Nicole, Sentient Foundation