The Policy Tester Interface is designed to run entirely on localhost. All AI inference is handled by a local Ollama instance — no policy text or analysis results are sent to any external service.

The application has no authentication, no user accounts, and no persistent storage. It is intended for local, single-user use during policy workshops or individual analysis.

• Policy text you enter stays on your machine. It is sent from the browser to the local FastAPI server, then to Ollama — all over localhost.
• Results are held in browser memory only and are lost when you close or refresh the tab.
• No telemetry or analytics are collected.

• Add external API calls that transmit policy text or results off the machine
• Add authentication or user management without a full threat model review
• Hard-code secrets, API keys, or tokens anywhere in the codebase
• Add persistent storage of policy text or results without considering data protection implications
• Expose any service on 0.0.0.0 — all services must bind to localhost only

• Keep all AI processing local via Ollama
• Validate and sanitise all user input on the backend (FastAPI/Pydantic)
• Pin dependency versions in requirements.txt and package.json
• Run npm audit and pip audit periodically to check for known vulnerabilities

• Frontend: Basic checks before sending requests (non-empty policy text, at least one category selected)
• Backend: Pydantic model validation on all API requests; invalid categories are rejected with a 400 error; empty policy text is rejected

1. No rate limiting — the FastAPI server does not limit request frequency. This is acceptable for localhost use but would need to be addressed before any network deployment.
2. No HTTPS — all traffic is over plain HTTP on localhost. Do not expose these services to a network without adding TLS.
3. No Content Security Policy — the Vite dev server does not set CSP headers. The production build should add appropriate headers if deployed.
4. LLM output is not sanitised for XSS — model responses are rendered as text in React (which escapes HTML by default), but if the rendering changes to use dangerouslySetInnerHTML, output sanitisation would be required.

If you find a security issue, please open a GitHub issue or contact the maintainers directly. This is a hackathon project and does not have a formal security response process.