Securing Open Source AI: The OML Framework

​About the Event:
The moment you release open-model weights, you usually lose control. But what if you didn't have to?

​We are proposing a shift toward AI-native cryptography. This talk explores the OML (Open, Monetizable, Loyal) framework—a new approach to giving model builders cryptographic-style guarantees of attribution and control, even when the model is open.

​Key Takeaways:

• ​🛡️ Defense: How model fingerprinting can enforce attribution.

• ​⚔️ Offense: How adversaries can currently shatter "robust" fingerprints (and what we need to do to fix it).

• ​🐺 Bonus: A look at the Werewolf Agents Tournament—evaluating AI deception and persuasion in adversarial settings.

​This is an essential talk for builders and researchers at the intersection of AI Safety, Cryptography, and Decentralized Tech.

​Speaker:
Edoardo Contente (AI Researcher @ Sentient | Princeton Alum)

​Join the AER LABS Network:
🌐 Website: https://aerlabs.tech/
✍️ Blog: https://aerlabs.tech/blogs
📺 YouTube: https://youtube.com/@aerlabs
💼 LinkedIn: https://linkedin.com/company/aer-labs
💬 Discord: https://discord.gg/cR9Vn9zR