Project ARA is a small, community-maintained tool. Security reports are taken seriously and handled on a best-effort basis.

The project is pre-1.0 and moves on main. Only the latest main is supported — please confirm an issue reproduces there before reporting.

Please do not open a public issue for a security vulnerability.

Use GitHub's private vulnerability reporting:

1. Go to the repo's Security tab.
2. Click "Report a vulnerability".
3. Describe the issue, the impact, and steps to reproduce.

This keeps the report private until a fix is available. Expect an initial response within about a week. If accepted, we'll work on a fix and coordinate disclosure; if declined, we'll explain why.

ARA runs locally and has no network service, no authentication, and no remote attack surface of its own. The areas most relevant to security are:

• Reading untrusted metadata. ARA reads config.json, Info.plist, package metadata, and Homebrew output from the local system. Parsing is best-effort and defensive, but a crafted file is the most plausible input-handling concern.
• Subprocesses. ARA shells out to read-only tools (brew, nvidia-smi, system_profiler, interpreters for --version/metadata). It does not pass untrusted input as shell strings, and recon never mutates state.
• Model downloads. ara profile can download a calibration model via huggingface_hub — only with explicit consent, into the standard HF cache.

If you find a way ARA could execute untrusted code, leak credentials (e.g. an HF token), or mutate the system without consent, that's in scope — please report it privately.