Deception Benchmark

An AWS evaluation set of code samples in which real vulnerability patterns sit alongside controls that block exploitation, used to test whether a model can tell exploitable code from safe lookalikes. Samples are public, labels are withheld, and predictions are scored by AWS.