Evidence-first AI governance RAG

PolicyProof

Ask a question about the controlled NIST, EU AI Act, and OpenAI system-card corpus. The portable demo retrieves passages, estimates evidence sufficiency, returns source-derived excerpts with citations, or abstains.

Ask the corpus


Important disclosures

Portable ranking: deterministic BM25 is used because the benchmark-selected dense model asset is intentionally not committed.

Evaluation: sufficiency results use construction-derived silver labels, not human-adjudicated gold labels.

Responsible use: this is research and compliance support, not legal advice or a compliance determination.