Proteus and ZK Email How Advanced Cryptography Derisks AI Harms
summary
Proteus is a system that combines DINO Hash perceptual hashing, digital signatures and privacy tech to verify image provenance; claims include orders-of-magnitude better matching than Apple’s neural hash and Facebook’s approach, over 5,000 real ZK signature proofs already produced, and adversarial training that preserves ~60% of hash bits under small pixel attacks. Private lookups run via MPC/FHE (WorldCoin–style clustered matching at billions/sec cited) and ZK Email provides signed inputs and selective disclosure; a simple AI detector (synthbuster) serves as a cold‑start fallback.
AI-written summary · about
chapters (22)
- 0:00Introduction: Cryptography’s Role in the AI Era
- 0:50The Problem: Verifying Content Provenance
- 1:49Why Watermarking Fails Against Deepfakes
- 3:21Digital Signatures as a Robust Alternative
- 4:49Real-World Use Cases for ZK Email
- 5:48How ZK Email Works: Privacy-Preserving Proofs
- 6:16Practical Applications (Docs, Flights, GitHub)
- 7:36User Adoption & Proof Volume
- 8:01Why Image Metadata-Based Provenance Fails
- 9:37Flaws in C2PA & Timestamp Vulnerabilities
- 10:06Introducing Perceptual Hashing
- 11:34Proteus: Combining Signatures + Perceptual Hashing
- 12:29How Perceptual Hashing Resists Modifications
- 13:59DINO Hash: A Better Perceptual Hashing Model
- 14:28Building a Decentralized Provenance System
- 15:58Privacy Challenges in AI-Generated Content
- 16:28MPC/FHE for Private Provenance Lookups
- 17:51Backup AI Detection for Cold Start
- 18:16Benchmarking DINO Hash vs. Existing Solutions
- 19:14Adversarial Robustness in Perceptual Hashing
- 20:11Final Takeaways & Future Work
- 20:40Resources & How to Get Involved
related talks
37mWhat ZK Can—and Can’t—Do for You
38mPrivacy Enhancing Cryptography at NIST
51mWhy you shouldn't trust me
16mTruth in a Post Truth World
23mHow to overcome AI threat to user authenticity
51mPanel: Zero Knowledge in the Enterprise
34mSNARKs and their Practical Applications
19mInnovations in Privacy Technologies: The Keys To Decentralized Networks