This research introduces SAFE, a framework that overcomes the lack of training data for automated proof generation by leveraging LLM self-evolution techniques.

• Creates a bootstrapping approach where LLMs progressively learn to generate formal proofs for Rust code
• Addresses the critical shortage of human-written proofs as training data
• Achieves verification of real-world Rust functions without requiring manual proof-writing
• Demonstrates significant improvement in automated security verification capabilities

This breakthrough matters because it reduces the enormous manual effort traditionally required for formal verification, making security guarantees more accessible and practical for software developers.

Automated Proof Generation for Rust Code via Self-Evolution