opening a model shouldn’t mean giving up the ability to prove it’s yours. that’s the problem we’re tackling at @sentient_found. this leaderboard tests how well model fingerprints survive attacks and how much model quality they preserve.

This is the missing piece for real open-source AI. Training costs millions; copying takes minutes. Embedding provenance directly into model weights, rather than relying on paperwork, means models can remain open without losing their origin or credit. The historical parallels are spot on, and an open leaderboard is exactly what this research needs to move forward. A big step for open-source AI. Kudos to the @SentientAGI for mmaking this possible.

as AI models become easier to copy, fine-tune, and distill, proving where they came from matters more than ever. renaissance painters left hidden artist marks on their masterpiece and now, open source ai creators and builders should have a way to mark the intelligence they build & prove where it came from. 🤗

模型训练可能耗资数百万美元，但复制它只需要几分钟 微调和蒸馏可能会抹去模型的来源痕迹 我们推出 Fingerprinting Leaderboard——一个用于测试模型来源证明的开放基准 我们的目标：让证明本身存在于模型之中 fingerprint.sentient.foundat… 模型来源的证明通常存在于模型之外的日志和元数据中 但这些痕迹可能随着复制、微调或蒸馏而消失 模型本身需要携带证明 这就是模型指纹技术 @SentientAGI Labs 与 @Princeton 的研究人员 @edoardocontente、@viswanathpramod、@sewoong79、@oleg_golev 和 @ZeruiCheng 在训练过程中嵌入秘密的查询-响应对（query-response pairs），以帮助识别模型的来源

AI distillation is a problem Fingerprinting is the solution Check out the fingerprinting leaderboard here:

Sentient is solving the model provenance to prove where it came from. We are here to support the Open-source AGI

• Lexicographers planted fake words in dictionaries. • Cartographers drew fake streets into maps. • Artists hid their signatures inside the paint. • AI builders embed fingerprints into their models. History doesn't repeat itself, but it does rhyme.

Lexicographers planted words that did not exist. Mapmakers drew streets that led nowhere. Artists placed hidden signatures inside the work itself. Long before AI, creators had ways to catch copying to prove provenance ↓

Lexicographers planted words that did not exist. Mapmakers drew streets that led nowhere. Artists placed hidden signatures inside the work itself. Long before AI, creators had ways to catch copying to prove provenance ↓

Had an amazing session with @Ewhachain at Ewha Womans University 🏫 Spoke to a room full of brilliant students about a question I keep coming back to: If AI models can be copied in minutes — who actually owns them? That's the problem @SentientAGI is solving. Covered a lot of ground in 90 minutes: → Why AI infrastructure is dangerously concentrated → How fingerprinting proves model ownership → What open-source AI actually looks like in practice (EvoSkill, Arena, ROMA) Huge thank you to everyone who showed up and asked such thoughtful questions : ) And stay tuned with @sentient_found 's upcoming updates!

The most important part of Sentient’s $42M isn’t the number. It’s the idea behind it. For open AI to truly compete with closed systems, good ideas alone aren’t enough. Developers and researchers need the capital, resources, and freedom to keep building. That’s what makes Sentient Foundation’s Open Source AGI Grant & Investment Program stand out. On the grant side, there’s no equity, no lockup, and no claim on the work developers create. That changes the equation. The goal isn’t simply to fund a few projects today. It’s to give the people building the future of open AI more room to experiment, create, and keep pushing the ecosystem forward. I don’t think the real impact of $42M will be measured by how much funding is distributed. It will be measured by what gets built because that funding existed. @SentientAGI @sentient_found

This is one of the most important open-model papers of the year by @SentientAGI 24,576 fingerprints hidden inside Llama-3.1-8B. That’s ~100× more than previous work, with barely any impact on model utility. Fine below the benefits of this: • Leak a few keys? Thousands still remain. • Just 5 queries can distinguish two models. • Even colluding hosts can still be detected. • After Alpaca fine-tuning, ~83% of 1,024 fingerprints survived. • The fingerprints look like natural language, not obvious gibberish. And they maintained >95% utility with 8,192 fingerprints across Llama, OLMo-2, Qwen-2.5, Mistral and Phi-3. This is the missing piece for open weights: You can release the model, let others fine-tune and deploy it, while still having a way to prove provenance from API access alone. Open weights don’t have to mean giving up attribution. That’s a huge unlock for open AI.