Boundary (AI Bug Bounty)

Boundary is a self-hosted platform that explores a specific question: how can AI agents help with a security assessment without being handed the keys? The usual worry with autonomous tooling is that it does too much, drifts outside what was agreed, or leaves no clear record of what happened. Boundary is designed around the opposite principle.

The guiding rule is the model proposes, the platform decides. Before any work begins, the engagement scope is defined, reviewed for ambiguity, and frozen as an immutable policy snapshot. From then on an agent can only suggest the next step; Boundary checks each suggestion against that approved scope and carries out the allowed ones through its own constrained gateway. Everything that is proposed, approved and performed is recorded, so the entire assessment can be audited afterwards.

It is intended strictly for systems the operator is explicitly authorized to assess, and the generated scope has to be reviewed and approved before anything runs. The project is really about governance and control for AI-assisted security work rather than about any single technique.