Evaluate the action
Examine the requested operation and its evidence against explicit security policies. Return an allow, block or approval-required policy decision.
AI AGENT ACTION SECURITY
Before agents act.
Make the call.
An agent can have valid credentials and still attempt the wrong action. DUSK evaluates what it wants to do, applies security policies and guards execution with signed permits.
Open-source project · Controlled integrations · Evidence before claims
01 / INPUTAgent actionA request to change something.
02 / DUSKEvaluate & authorizePolicy decision. Signed permit.
03 / EXECUTORVerify before executionNo valid permit, no action.
THE CONTROL BOUNDARYPermission is a starting point.
Permission is a starting point.
Every action still matters.
Guard the execution
Use signed permits at the executor boundary. Protection depends on routing the relevant actions through that boundary.
Keep the evidence
Record decisions and audit evidence so operators can investigate what was requested, what was decided and what happened next.
BUILT IN THE OPENInspect the work.
Inspect the work.
Understand the boundaries.
DUSK brings together action policy, permit verification, audit services and an operations console. Development and testing take place in controlled environments.
This is the public project website. Hosted customer access is not available here yet. Live integrations and production readiness are being qualified separately.