Please report security issues privately, not as a public GitHub issue.

Use GitHub's private vulnerability reporting (the "Report a vulnerability" button on the Security tab). That opens a private thread visible only to you and the maintainers.

Please include what you did, what happened, and what you expected. A short reproduction is worth more than a long description.

This is a small project maintained by one person, so response time is best-effort rather than a guarantee. Expect an acknowledgement within about a week. If a report is valid and I cannot fix it quickly, I would rather say so than leave you waiting.

OutLoud is a local dictation tool. To do its job it holds two macOS permissions that deserve scrutiny, because they are the same ones a keylogger would want:

• Input Monitoring lets it see the hotkey. Without it the key never registers and nothing happens at all.
• Accessibility lets it read the focused text field and write into it. Without it text is delivered by clipboard paste instead of inserted in place.

Both are requested through the normal macOS permission system, and both can be revoked in System Settings at any time. You can ask the app what it currently holds:

OutLoud.app/Contents/MacOS/OutLoud --permissions

There is no network code. The binary does not link CFNetwork or any other networking framework, which you can check yourself on a build you compiled:

otool -L OutLoud.app/Contents/MacOS/OutLoud | grep -i network

Speech recognition runs on-device through Apple's SpeechTranscriber. No audio, transcript, or telemetry is uploaded, because there is nothing in the binary capable of uploading it.

Diagnostics are user-initiated only. scripts/doctor.sh --report prints a redacted section intended for pasting into an issue; it is generated locally and nothing is sent anywhere.

If you are reviewing this, these are the parts where a mistake would matter most:

• Text injection (crates/outloud/src/inject.rs). Chooses between accessibility writes, synthetic keystrokes, and clipboard paste. A routing mistake here can deliver text to a window the user was not looking at.
• Clipboard handling (paste_with_leading_space). The clipboard is saved and restored around a paste. A failure to restore would silently destroy whatever the user had copied.
• Shell bridge (crates/shell-bridge/). A Unix domain socket that lets a shell pull a staged utterance, so dictation can edit a command line. The socket is created 0600 inside a 0700 directory, both asserted by tests, but it is the only IPC surface in the project and the most interesting thing to attack.
• Focus handling. The app records which application was focused at key-down and warns when focus moved before the text was written. Silent misdelivery is the failure mode this exists to prevent.

These are known and documented, so please do not spend your time on them:

• The app is ad-hoc signed. It has no Apple Developer ID, so macOS Gatekeeper will warn on first launch and every rebuild silently voids the accessibility and input-monitoring grants. This is a funding gap, not a flaw, and it is why the project currently expects you to build from source.
• Dictated text goes to whatever has focus. That is the feature. The focus-moved warning exists because it can surprise you, not because the behaviour is wrong.
• The recognizer is Apple's. Accuracy problems, and anything about how SpeechTranscriber handles audio internally, belong to Apple rather than here.

The project is pre-1.0 and only the main branch is supported. There are no backports to older tags.