Revolutionizing Cryptography: AI Agents Automate Lean Formal Verification

Automating Lean Formal Verification of Cryptographic Protocols with AI Agents

The Nyx Foundation, located in Bunkyo-ku, Tokyo, has announced a groundbreaking achievement in the field of cryptography: the successful automation of Lean formal verification for cryptographic protocols using AI agents.

Understanding Lean Formal Verification

Formal verification of cryptographic protocols ensures the correctness of specifications and the security of cryptographic methods through mathematical proofs. Traditionally, the safety of cryptographic protocols has relied on human experts to perform these proofs. However, as protocols become more complex, the likelihood of human error increases, sometimes leading to critical flaws being identified years after a protocol was deemed secure.

In recent years, the difficulty of formulating and understanding these proofs has risen, even for experts. Incorporating formal methods allows for a more rigorous description and validation of these proofs. The AI agent developed by Nyx Foundation leverages a unique workflow to generate Lean language, automating the formal proofs of security claims in cryptographic specifications.

Achievements with the AI Agent

The AI agent has demonstrated its capabilities by successfully proving the formal security of a specific encoding scheme within the XMSS hash-based signature scheme. This result emphasizes the potential of AI in handling the complex task of formal verification.

For those interested, the implementation can be found in the following repository: GitHub - Nyx Foundation.

Formally verifying cryptographic protocols typically requires highly skilled personnel and significant time investment. The Nyx Foundation has effectively automated many of these processes by utilizing existing AI models and developing a unique workflow, thus streamlining a traditionally arduous task.

The Proposed Pipeline

The research introduces a pipeline utilizing AI agents for Lean formal verification of cryptographic protocols, which includes the following components:
1. Natural language generation for non-formal proof drafting.
2. Review processes by a checklist tailored for cryptographic verification, determining the acceptance of proofs.
3. Automatic transformation of accepted drafts into Lean language for type checking and proof verification.

This double-gate design was adopted to enhance both efficiency and quality in verifying formal claims. Evaluation metrics, such as the success rate of Lean verification on various prototypical propositions, serve to measure the efficiency and quality of results.

While several papers have been published on automating Lean formal proofs using AI agents, very few have attempted to apply this to cryptography. As of now, no systematic reports exist in the literature regarding the application of Lean and AI agents within the domain of cryptographic protocols.

Future Perspectives

Scheduled for November 2025, the Nyx Foundation's research will be showcased at the 24th International Conference on Cryptology and Network Security (CANS2025) as a poster presentation. Those interested in this innovative work are encouraged to attend.

Additionally, the AI agent will be open-sourced, with plans to improve its accuracy continuously. The Nyx Foundation aims to share more information progressively as development continues.

Harnessing their deep expertise in cryptography and formal verification alongside advanced software engineering, the Nyx Foundation is at the forefront of automating formal verification methodologies using AI agents.

About Nyx Foundation

Nyx Foundation is a non-profit research institution based in Bunkyo-ku, Tokyo, focusing on the Ethereum blockchain. Financial support comes solely from donations, grants, and sponsorships, enabling the foundation to address significant challenges in the Ethereum ecosystem. Collaborations have already begun with the Ethereum Foundation and various blockchain enterprises, including notable presentations at institutions like Columbia Business School.