The Coinbase exchange announced on April 14, 2026 the development of Frosty, an artificial intelligence tool designed to audit smart contracts, which is already part of its internal security review processes.
According to the companythe system was evaluated alongside six other AI tools in a process that analyzed 33 real audits with 434 verified vulnerabilities. In these tests, Frosty obtained better results in metrics such as precision, coverage and F1 score, used to measure effectiveness in detecting faults.
The tool works through an autonomous architecture based on multiple agents and sequential phases. The process includes tasks such as code analysis, searching for vulnerabilities, adversarial reasoning—to simulate possible attacks—, debugging findings, and generating preliminary reports. Each execution takes between one and two hours and produces a report that is subsequently reviewed by human teams.
According to the company, The incorporation of this type of systems must be to the growing use of artificial intelligence both by developers and potential attackers. In this context, automated tools seek to accelerate the detection of errors in early stages of development.
However, Coinbase clarifies that Frosty does not replace traditional audits carried out by specialists. The tool can overlook complex or contextual vulnerabilities, so its use is intended as a complement to the review process.
The development of solutions of this type occurs in parallel with other initiatives in the sector. For example, OpenAI recently introduced EVMbencha testing environment to measure the performance of artificial intelligence agents in the detection, correction and exploitation of faults in smart contracts, as reported by CriptoNoticias. These tools have shown progress, although with uneven results depending on the task.