Skip to content

Security: debe/borg_ai

Security

SECURITY.md

Security

Reporting a vulnerability

Report it privately through GitHub: the repository's Security tab, then Report a vulnerability. Do not open a public issue for it. If private reporting is not available, open an issue that asks for a private contact and leave out the details.

Please include:

  • the borg version (the release, or cargo install --list) and the operating system;
  • which part is affected (the hooks, borg-mesh, borg-web, an adapter, the relay setup);
  • the steps or a test case that shows it, and what an attacker gains.

We aim to answer within a week. Fixes go into the next release; the changelog names the issue once a fixed release is out.

Supported versions

Only the newest release gets security fixes.

What counts

Borg handles three kinds of trust:

  • Your machine. Hooks run as you, with your agent's input. A way to make borg run a command, write outside its stores, or read a repo routed off is a vulnerability.
  • The mesh. Membership is a member list signed by an admin, every lesson line is signed by the machine that wrote it, and steering directives are checked against the member list. A way for a non-member or a revoked key to join or inject lines, or for a member without the steer role to steer drones, is a vulnerability. So is lesson or presence content reaching a relay or n0 in clear text.
  • What leaves the machine. Raw episodes, transcripts, local collectives and off repos never sync. Anything that makes them travel is a vulnerability. A secret that survives redact in a reasoning excerpt or in presence is a bug worth reporting privately too.

Not in scope: what a relay sees by design (node ids, IP addresses, traffic volume; see Running your own relay), what the model provider receives for a model step (Privacy), and lessons that quote a secret the model wrote into them (read lessons.jsonl before you share a collective that handled secrets).

There aren't any published security advisories