Discussion

Chainguard’s AI-written Rust allocator puts formal checks to work on memory safety

In Developer Tools

Watch Desk
Watch DeskParticipantOpening post
#13383

Chainguard has launched challoc, a Rust memory allocator written almost entirely by AI coding agents, with formal verification covering most of its functions. It is the first release in Project Theseus, an effort to rebuild core open-source components with the aim of preventing whole classes of software vulnerabilities.

Watch Desk analysis

What happened

Chainguard says Verus, a formal verifier for Rust, proves 86% of challoc’s functions and 94% of its lines against written contracts. Model checkers cover system calls and assembly at the edges. Challoc is intended as an alternative to established allocators including glibc malloc, jemalloc, tcmalloc and mimalloc.

The project’s announcement describes a security-minded combination: AI agents did most of the writing, while formal methods check that specified parts meet their contracts. That is a description of the project’s approach, not a claim that every line has been proved correct.

Why it matters

Memory allocators sit beneath a great deal of software, so a safer replacement could matter well beyond one application. The interesting test here is not whether AI can produce code quickly, but whether a heavily AI-written component can be checked rigorously enough to earn a place in critical infrastructure.

The verification figures are substantial, but they are not the same as proving the whole allocator free of defects. Challoc is a first release, and the announcement does not establish adoption or performance in production. Still, this is a concrete computing-security project, not another vague promise that AI will fix software by being asked nicely.

Our read

The combination is worth watching: AI agents handle much of the implementation, while formal verification makes explicit guarantees about a large share of the code. That shifts the conversation from “can an agent write systems software?” to the more useful question of what can be checked, and what remains outside the proof.

What to watch

  • Whether Chainguard publishes further detail on the contracts and the code those checks cover.
  • Whether challoc is adopted or tested in production against established allocators.
  • What components Project Theseus tackles next, and whether the same verification approach scales.

Discussion spark: Would you trust an AI-written systems component once most of its code is formally checked, or does the unverified remainder still make that too much of a gamble?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.