Canonical and University of Bristol Use AI to Translate Legacy C Code to Rust
A new research partnership aims to automate the migration of security-critical Linux components to Rust using a neurosymbolic verification system.
Canonical has partnered with researchers at the University of Bristol to determine if AI and automated tools can reliably rewrite legacy C code into safe, maintainable Rust. The project seeks to move beyond simple code generation by creating a system that proves the translated code is behaviorally identical to the original.
To test the system, the partnership is focusing on security-critical components including AppArmor, a Linux security module, and snap-confine, which manages sandboxed environments for snaps. The proposed workflow utilizes a neurosymbolic approach: language models handle the initial translation, while a rigorous verification process—combining fuzz testing and formal program analysis—detects behavioral differences between the C and Rust versions. When the system identifies a mismatch, it employs symbolic program-repair techniques to diagnose the failure and fix the code.
The Memory Safety Gap
Much of the world's critical infrastructure and operating systems remain written in C, a language prone to memory-safety vulnerabilities such as buffer overflows and use-after-free bugs. While the Rust programming language provides native protections against these flaws, the sheer volume of legacy C code makes manual rewriting prohibitively expensive and risky.
Automated translation via AI offers a potential shortcut, but it introduces a new set of risks. Generated Rust code can compile cleanly while still behaving differently from the original C code. This creates a trust gap for maintainers, as code that passes a compiler may still contain subtle logic errors that compromise production systems.
Eliminating 'Unsafe' Rust
A primary technical hurdle for the project is ensuring the resulting code is actually safe. A naive translation often results in the overuse of 'unsafe' Rust blocks, which essentially bypass the language's safety guarantees. If the AI simply reproduces C syntax within these blocks, the translated code carries over the same memory-safety risks present in the original C source.
The goal of the Bristol and Canonical project is to produce idiomatic Rust that avoids these pitfalls, ensuring the migration provides a genuine security upgrade rather than a superficial syntax change.
Implications for Infrastructure
If the partnership can prove that AI-generated Rust is both memory-safe and behaviorally identical to legacy C at scale, it could trigger a massive migration of critical infrastructure to safer languages. Such a breakthrough would allow the Linux ecosystem to eliminate entire classes of security vulnerabilities without requiring decades of manual labor.
Observers will be watching whether the system can handle the complexity of AppArmor and snap-confine without introducing regressions. The success of these industrial case studies will determine if automated translation is a viable path for the broader operating system community.