Canonical is co-funding a three-year PhD research project at the University of Bristol to determine whether artificial intelligence can successfully convert massive C codebases into Rust. The company's engineering vice president, Jon Seager, announced the investment on Ubuntu's Discourse forum, stressing that this isn't a plan to immediately rewrite Ubuntu with AI-generated code. Instead, Canonical will finance a doctoral candidate to spend several years examining whether the concept is technically feasible, producing evidence about the approach's viability in an industry saturated with overblown claims.

The research will focus on programs "comprising hundreds of thousands of lines of C" and attempt to break them down into smaller pieces before deploying an LLM to rewrite those pieces in "safe, behaviourally correct and maintainable Rust," according to Seager's announcement. The project will specifically examine two tools: snap-confine and AppArmor. Canonical is co-sponsoring the work with UK Research and Innovation, a public agency sponsored by the UK's Department for Business, Innovation, Science and Trade. The University of Bristol's Professor Meng Wang and Dr. Cristina David will oversee the effort alongside Seager.

Seager's post runs just over 1,000 words and tackles several likely objections head-on. Traditional source-to-source translation tools can handle substantial volumes of code, but they "often preserve the structure of the C too literally," the announcement notes. The output might compile as Rust, yet still depends heavily on unsafe operations, keeps awkward C conventions, and demands considerable manual effort before it looks like code a Rust developer would want to maintain. The proposal acknowledges that established codebases hold knowledge their original programmers never explicitly documented—years of bug fixes and patches that encode responses to real-world edge cases nobody foresaw at the start. Machine translation might retain some of that behavior, while a fresh human rewrite based on the original design could miss it entirely.

The most difficult challenge won't be translating the code itself, but rather decomposing a large codebase into smaller components that automated systems can process, the report suggests. This recalls decades of research trying to divide arbitrary algorithms automatically into tasks that can be distributed to parallel processes—work that has yielded useful techniques for specific situations but no universal solution. The report expresses staunch skepticism about generative AI outside narrow domains like human language translation, noting grave doubts about viability and speculating the problem may prove incomputable, similar to the Halting Problem. Yet the three-year timeframe matches conventional UK PhD programs and should produce the evidence the industry needs. As the report speculated in 2024, automatic translation between programming languages could become immensely valuable for improving software reliability—not by fixing problems automatically, but by exposing previously unknown errors. For organizations still relying on AppArmor rather than the more widely adopted SELinux in enterprise Linux, a hardened Rust implementation could deliver meaningful security benefits, particularly since openSUSE replaced AppArmor with SELinux last year. The decision to fund rigorous academic research rather than immediately deploying AI tools represents a measured approach that prioritizes understanding over hype, with results expected when the doctoral work concludes. If the university's systematic investigation confirms that decomposition and translation can be automated reliably, the tooling could reshape how organizations approach technical debt in aging but critical infrastructure. Whether the PhD candidate emerges with proof of concept or proof of impossibility, enterprises will finally have empirical guidance on whether to invest in this approach or pursue alternative modernization strategies.