02/09/2026
"An audit is too big for one prompt." ๐บ
In 2026, when LLMs are producing a major portion of the code that ends up in the repository, our Head of Web Development, Victor Motogna, has written an explanation of how fintech code audits actually function.
He begins by saying that the responsibility for audit-readiness remains with the development team, since AI is responsible for writing more of the code.
What works is a loop of small steps where each one is either deterministic or LLM-driven. An LLM flag is a lead, and it becomes a finding once a failing test or a rerunnable command backs it up.
Victor walks through all seven phases, the C1-C15 check pack, the majority-vote review runs, and the last mile only a person can walk. ๐
How to audit fintech code when LLMs write much of it. Split the audit into static checks and narrow LLM reviews, and reproduce every flag before it counts.