Gene Lens is a local-first tool that handles your own genetic data on your own machine. Security and privacy are the same conversation here — if something can leak data or be coerced into doing so, treat it as a security issue.

If you discover a vulnerability — especially one involving genetic data handling, privacy leakage, or NetworkBlocker bypass — please do NOT open a public issue.

Instead, report it privately through either:

1. GitHub Security Advisories — the "Report a vulnerability" button on the Security tab. This is the preferred channel because it tracks the disclosure timeline and lets you collaborate on the fix in a private fork.
2. Email — the maintainer email listed on the profile. Subject line: [gene-lens security] <short description>.

• NetworkBlocker bypass — anything that causes data to be transmittable during analysis, including via subprocess, os.system, or dependency-level escape hatches.
• Path traversal — read/write outside the project's working directories from web routes (/analyze, /history/load/<hid>, /api/chat/ask, etc.).
• Genotype or PII leakage — through logs, error messages, generated reports, response metadata, or environment variables.
• CSRF / cross-origin — bypass of the before_request Origin/Referer check.
• Unsanitized model output — prompt injection that makes the AI chat exfiltrate context or execute embedded HTML/JS in the dashboard.
• Insecure deletion — uploaded files or chat transcripts left recoverable after the documented secure_delete / clear() calls.
• Dependency vulnerabilities with practical impact on this project's threat model (not every CVE qualifies — a CVE in Flask's debug toolbar doesn't matter for a tool that only listens on 127.0.0.1, for example).

Gene Lens is single-user, offline-first. The security model assumes:

1. The host machine is trusted. If your machine has malware, no software-level guarantee survives. We don't protect against dtrace, kernel module exfil, screen recording, or memory inspection by another process running as the same user.
2. The working directory is not cloud-synced. Dropbox, OneDrive, iCloud, Google Drive sync folders are out of scope — files placed in input/ or output/ will be uploaded by that software, not by us. The README calls this out explicitly.
3. Flask listens only on 127.0.0.1. If you front the dashboard with a tunnel (ngrok, tailscale, etc.) to make it reachable from elsewhere, that traffic path is your responsibility and outside our threat model.
4. The user runs download_databases.py deliberately. This is the only script that touches the network. Analysis paths assume the databases are already on disk.

The NetworkBlocker protects against accidental Python-level network calls during analysis. It does not protect against:

• Subprocesses started by future code (use subprocess carefully and audit each call against the analysis hot path)
• C extensions that bypass Python's socket layer
• OS-level networking from outside the Python process

If you find a way to coerce data out of the analysis hot path — even indirectly — that's in scope for this policy.

The project is pre-1.0. Only main is supported; security fixes won't be backported to earlier commits because there are no released tags yet. After v0.1.0 is cut, the most recent minor version will receive fixes.

Most of the safety floor is set up at install time. Quick checklist:

• Run python main.py privacy-check after every privacy-sensitive change you make to the repo.
• Keep the project directory outside any cloud-sync folder.
• Don't expose port 5000 past localhost. The Docker image's compose file binds 127.0.0.1:5000:5000 explicitly.
• Verify uploads are deleted: after an analysis finishes, input/ should not contain your DNA file (secure_delete runs automatically; if it doesn't, that's a bug worth reporting).
• For the optional AI chat: only install Ollama models you trust. The chat function feeds your analysis context to the model — that's the whole point — but if you've installed a malicious model, it sees that data too. Pull from the official Ollama library only.