Researchers develop GoCoMA tool to identify which AI wrote a piece of code—addressing security and licensing risks as LLM-generated code becomes indistinguishable from human work

Researchers published GoCoMA, a system that analyzes code to determine whether it was written by a human or generated by an LLM (large language model—AI trained on millions of code samples). The system examines two layers: the code's writing style and structure, plus the binary 'fingerprint' left by the compiler (the software that translates code into machine instructions).

Unlike previous detection methods that focus on one aspect of code, GoCoMA fuses multiple signals together using hyperbolic geometry (a mathematical technique that better captures hierarchical relationships), making it more accurate at catching LLM-generated code that mimics human writing. This matters because LLMs now write code so convincingly that existing detection methods struggle.

For business and security teams, this unlocks code auditing at scale—you can now scan open-source repositories or internal codebases to flag AI-generated contributions, catching potential licensing violations (using GPL code without attribution), supply-chain risks, or unvetted AI code before deployment. For companies managing compliance or intellectual property, this removes uncertainty about code origin.