CryptoDetector
Menu

Review the Specification Before Trusting AI-Generated Vault Code

For anyone relying on a vault contract to safeguard assets, the question is not only whether its code has been reviewed, but whether the review checks the right properties.

An article published on November 17, 2026, argues that AI-generated code can make code review a bottleneck. It proposes a different order of work: write the properties a change must satisfy in plain English, have people review those properties, and then use formal verification tools to check the code against them.

The article describes a demonstration by Certora CEO Mooly Sagiv. An agent inferred properties for an ERC-4626 USDC vault and generated Solidity code and CVL rules. A prover then found a dilution flaw of about 5 wei caused by high-water-mark rounding; the flaw was subsequently fixed. The demonstration is not described as an exploited loss.

The remaining challenge is deciding whether the specification itself is good enough. A prover can check code against stated rules, but that makes human review of those rules an important part of the process—not a step to skip.