Microsoft Research's PDD Kit formally verifies agent-written Rust login code
Overview
Microsoft Research describes PDD Kit, a research prototype that turns a developer's intent into a Verus formal specification and then proves that agent-written Rust code meets it for every input.
The method is demonstrated on the login logic of Pet Store, a full-stack Rust web app, where Verus uncovered existing bugs.
The proof does not reach every component. The bcrypt password check and database lookups are covered by tests and an explicitly accepted trust boundary instead. Microsoft says the same process could verify the functional correctness of the entire app, and PDD Kit currently supports only Rust because Verus works only on Rust.
Written by AI from the articles below · updated Nov 5, 2:03 PM ET
Check the sources:
Developments
2 developments
- Nov 5, 1:36 PM ET · 1 articleMicrosoft Research explains proof-driven development for AI-generated codeMicrosoft Research: Microsoft Research explains Proof-Driven Development for trusting AI-generated code
- Nov 5, 11:00 AM ET · 1 articleMicrosoft explains formal verification of agent-written web apps with PDD KitMicrosoft Research: Microsoft shows how PDD Kit formally verifies agent-written login code
Article timeline
The articles in this story. Times are ET.
Microsoft Research@MSFTResearchOfficialMicrosoft Research explains Proof-Driven Development for trusting AI-generated codeAIMicrosoft Research describes Proof-Driven Development Kit, which turns plain-language developer intent into a specification in Verus and applies formal methods to check AI agent-generated code. PDD Kit uses mathematical proofs to test code against the spec where possible and indicates where the proofs cannot reach.

- Microsoft ResearchOfficialPickMicrosoft shows how PDD Kit formally verifies agent-written login code
AIMicrosoft Research describes PDD Kit, a toolkit that turns a developer's intent into a Verus formal specification, then proves that agent-written Rust code meets it for every input. The post demonstrates the method on the login logic of Pet Store, a full-stack Rust web app, where Verus uncovered existing bugs. Components that a proof cannot reach, such as the bcrypt password check and database lookups, are instead covered by tests and an explicitly accepted trust boundary.
Heat Trend
Not enough continuous observations to show a trend yet.