A benchmark for evaluating AI/LLM agents at reverse-engineering encryption
protocols used by Android mobile applications. Each task provides a decompiled
Android app and captured network traffic; the goal is to identify the cipher,
extract the key material, and produce a working encryption implementation that
can decrypt the captured ciphertexts.
benchmark/
├── common/ # Shared resources (not a task) for all runs
│… See the full description on the dataset page:
https://huggingface.co/datasets/wirebench-neurips/wirebench.