Boundary adherence
Does the answer stay inside the information, actions, tools, and role the user already has?
Not more access. More useful AI inside the access already granted.
Classic access control decides what the system may retrieve. BSL Access Boundary measures whether the AI behaves inside that boundary.
The module does not grant access. It evaluates AI behavior against the boundary that already applies to the user.
Does the answer stay inside the information, actions, tools, and role the user already has?
Classifies outputs as LOW, MEDIUM, or HIGH based on inference, overreach, pressure, and tool misuse.
When possible, proposes a safer answer that stays helpful inside the user's permitted role.
Preserves structured evidence for security, compliance, and internal review.
A nurse may see medication and monitoring instructions, but the AI should not infer a hidden diagnosis from medication alone.
BSL Access Boundary sits above existing access control. The customer system remains authoritative for identity, permissions, RAG filters, and tool access.
Current commercial boundary: paid constrained pilot. The module is not production enforcement and does not replace customer access control.
Start with one role, one workflow, and customer-approved synthetic cases.