Interest in math proof of code is more than firms can supply
Interest in formal verification is more than the supply, and AI written specifications can cause trust that is not right.
Formal verification is a math check that code does the work that its authors want. Proofcraft, a firm in this field, tells that many inquiries from customers are more than it and its competitors can supply. Startups try to use AI to write the specification, which describes how a software tool must work. 2 cybersecurity executives tell that they could pay AI labs for verified AI code.
How it works
AI writes the specification. An example is that a data table holds the same items before and after the tool sorts it. More tools then change it into a shape that proof tools can check.
The proof
In 2026, the security team of Apple told that formal verification found issues in its corecrypto encryption library that no other method could find. Apple adds new code carefully, to keep pace with the inspection. No count of inquiries is given. The executives work at companies that have not verified code. They are not sure that automation helps enough.
What is not known
Proofcraft leaders think that AI written specifications can cause unjustified trust. Maintainer Michael Rosenberg does not know who will pay for more checks of open source code. Some startups do not know if they must sell to users or through labs.
Insight: Whoever writes the specification defines verified
In the InfraDev test, probes reject 32% of code changes that the first tests accept: a good result is only as good as the check. Verification has the same limit when AI writes the specification. This reads as a new market for AI labs: they write the code, write the specification and sell the proof. The buyer must then check who wrote the specification.
Sources
Posted
