Description-Behavior Mismatch
Medium
- Confidence
- 93% confidence
- Finding
- In benchmark mode, the script prints example text snippets pulled directly from the memory database. That creates an unintended data-disclosure path: anyone running the benchmark against a sensitive RAG or memory store may expose private memory contents to logs, terminals, or calling systems, even though benchmarking search quality does not require revealing raw text.
