---
license: apache-2.0
pretty_name: PolicyBench
language:
  - en
tags:
  - clinical-ai
  - llm-security
  - ai-agents
  - healthcare
  - haarf
  - quokkaguard
  - synthetic
size_categories:
  - n<1K
task_categories:
  - other
---

# PolicyBench

**Context-aware, auditable policy-as-code authorization for clinical AI agents, with Cedar and Rego behind one trait.**

PolicyBench is the benchmark corpus for *Policy-as-Code Authorization for Clinical AI Agents: Context-Aware Decisions and a Cedar-vs-Rego Comparison*, the `quwarden` module of the QUOKKAGUARD program. It ships with the [`quwarden`](https://github.com/quome-cloud/quwarden) repository, which contains the `qfire` gateway layer under test, the experiment harness, and the paper.

Semantically equivalent clinical authorization policies in Cedar and Rego, a synthetic PIP attribute store, a HAARF coverage map, and 56 labeled request-to-decision cases (seed 42) covering the happy path, out-of-range dose (confused deputy), no-active-encounter, unpaneled patient, and role edges.

**All data are synthetic.** No real patient data or protected health information (PHI) is included; clinical content is generated from templates with fixed seeds.

## Files

| File | Size | Rows |
|---|---|---|
| `attrs.json` | 184 B | |
| `cases/cases.jsonl` | 11 KB | 56 rows |
| `cedar/clinical.cedar` | 796 B | |
| `coverage-map.json` | 246 B | |
| `rego/clinical.rego` | 815 B | |

### Record schemas

- `cases/cases.jsonl`: `action`, `args`, `attrs`, `expect`, `principal`, `resource`

## How it was generated

The corpus is produced by the generator in the paper repository and is fully deterministic (fixed seeds), so it can be regenerated byte-for-byte.

```bash
# from the quwarden repository root (deterministic seeds)
python3 scripts/005-policy-authz/gen.py --out datasets/005-policy-authz/cases
```

## Intended use

Evaluating the **policy** enforcement layer of a clinical-agent security gateway (HAARF control C8): A static tool allow-list lets an agent order opioids as long as the role permits it, ignoring dose, encounter, and panel: on PolicyBench, context-blind RBAC over-permits 42.9% of requests.

The experiments that consume it (E-series in the paper) are reproduced from the repository:

```bash
git clone https://github.com/quome-cloud/quwarden
cd quwarden
cargo build --release
```

then follow the *Reproduce the experiments* section of its README.

## Citation

This benchmark was built to evaluate a control of the **Healthcare AI Agents Regulatory Framework (HAARF)**, the source framework for the QUOKKAGUARD program. Please cite both the paper and HAARF:

```bibtex
@unpublished{schwoebel2026quwarden,
  author = {Schwoebel, James},
  title  = {Policy-as-Code Authorization for Clinical AI Agents: Context-Aware Decisions and a Cedar-vs-Rego Comparison},
  note   = {Preprint. Quome, QUOKKAGUARD program (quwarden module)},
  year   = {2026},
  url    = {https://github.com/quome-cloud/quwarden}
}

@unpublished{schwoebel2026haarf,
  author = {Schwoebel, Jim and Frasch, Martin and Spalding, Art and Sewell, Ed and Englert, Phil and Halpert, Ben and Overbay, Collin and Semenec, Ingrida and Shor, Joel},
  title  = {{HAARF}: Healthcare {AI} agents regulatory framework --- a comprehensive security verification standard for autonomous {AI} systems in clinical environments},
  note   = {medRxiv Preprint},
  year   = {2026},
  month  = {April},
  doi    = {10.64898/2026.04.09.26350519},
  url    = {https://www.medrxiv.org/content/10.64898/2026.04.09.26350519v1}
}
```

## License

Apache License 2.0. Copyright (c) 2026 Quome, Inc.