Meet the team building the universal verification layer for frontier AI labs and critical infrastructure, at the intersection of AI, security and cryptography.
Attestable is the only true zero-trust solution for AI integrity. Zero-knowledge proofs that guarantee AI systems are safe, private, and tamper-resistant.
Anthropic’s Claude team advances **provable inference** using zkSNARKs, enabling users to cryptographically verify AI responses without exposing model weights. Their **FlashProve** system generates proofs at **110 tokens/second** on a single H100 GPU for a 400B-parameter model (Llama 4 Maverick), with **~19 MB proof size per 1,024 tokens**. The open-source release includes a **C++ verifier**, **Lean-formalized proofs**, and a **technical report**, aiming to accelerate trust in AI outputs for security, regulatory compliance, and frontier pacing. Applications span consumer integrity checks and auditor verification of model behavior.