Please report security issues privately through GitHub's private vulnerability reporting on this repository, not as a public issue.

Helpful to include: the LockPlus version (Settings › About), the macOS version, what you observed, and steps or a profile (with secrets removed) that reproduce it.

What happens next:

• I acknowledge your report within 7 days.
• I keep you informed while I investigate and fix it.
• Fixes ship as a signed update through LockPlus's built-in updater; the advisory is published after the fix is out.
• I'd like to coordinate disclosure with you; the default is publication once a fix is released, or after 90 days, whichever comes first.
• With your permission, you are credited in the advisory and the release notes.

LockPlus is maintained by one person, so please be patient; I take every report seriously.

In scope:

• the LockPlus app and its network (system) extension;
• the app ↔ extension XPC interface;
• handling of profiles, keys and passwords (keychain, profile files, logs, the diagnostics export);
• the kill switch and DNS handling, beyond the limits documented in the README FAQ;
• the update chain (Sparkle feed and signatures);
• the migration from LockPlus 1.x and the removal of its old helper.

Out of scope, or better reported elsewhere:

• vulnerabilities in OpenVPN 3, wireguard-go or OpenSSL themselves: please report them to those projects as well (LockPlus ships them pinned and updates when fixes are released);
• macOS behaviour that LockPlus documents as a limit, e.g. the short window after a tunnel-process crash or the traffic macOS always lets bypass a VPN (see "How tight is the kill switch?" in the README);
• issues that need an already compromised Mac or administrator access;
• your VPN server's configuration.

LockPlus has not had an independent human security audit. Since 2.0.1, every security-relevant change is reviewed by two independent AI review engines from different model families: Anthropic's Fable and OpenAI's Codex (GPT-6-Astra at maximum reasoning effort). They review separately against the same threat model; findings are fixed and both review again until neither reports a high-severity issue. The reports and their fixes are in docs/reviews/. Two engines catch more than one, but they are not a human audit: independent review is very welcome.