AI-generated Rust compiles perfectly. That’s the scary part.

Canonical is asking a very specific question: can automated tools turn legacy C into Rust without changing what the software actually does? To test that, researchers at the University of Bristol are aiming the idea at AppArmor and snap-confine, two security-critical parts of Canonical’s stack.

AppArmor controls what applications can access, while snap-confine helps build the sandboxed environments where snaps run. Canonical is not trying to rewrite those systems by hand in Rust. It wants evidence that a machine-generated rewrite could be trusted in production.

The plan is to use language models to produce Rust, then push that code through a verification pipeline. The researchers want fuzzing and formal program analysis to compare the new version against the old C and catch differences that normal tests might miss. If something doesn’t match, symbolic repair is supposed to track down the failure and fix the code directly.

That matters because a Rust file can compile cleanly and still behave differently. And in this case, “different” can mean broken policy parsing, a bad sandbox setup, or a security hole where one didn’t exist before. Rewriting C in Rust can eliminate bugs like use-after-free and buffer overflows, but only if the translator doesn’t drag the old mistakes along through heavy use of unsafe.

That is the real problem Canonical and Bristol are trying to solve. Not whether AI can spit out Rust. Whether maintainers can trust it across huge codebases with hundreds of thousands of lines, where the hard part is proving the new code matches the old one closely enough to replace it.