BaxBench

Record description

BaxBench evaluates generated backend implementations using functional tests and expert-designed security exploits. Its task construction spans application scenarios and development frameworks, so a successful API implementation can still fail the associated security evaluation.

It is useful for analyzing the gap between feature completion and exploit resistance. Its reported results apply to the benchmark's tasks, prompts and tested models, not directly to production incident rates. The project is distinct from a deployed application builder and from the security-training approaches that later use it for evaluation.

Explicitly related research (2)

Declared connections