Research
A falsifiable design specification and verification program.
TenantSage treats governance claims as hypotheses that should be testable against explicit invariants, threat models, canonical artifacts and commit-bound conformance evidence. The goal is not to market trust, but to make governance behavior observable and disprovable when it fails.
Preprint
Beyond Brittle ACLs
Published in September 2026 by Arthit Pukhampuang, Beyond Brittle ACLs: Deterministic Governance and AI Visibility Boundaries in TenantSage describes a proposed S0–S8 control architecture. It is a design specification. Its publication does not establish that every stage has been implemented, independently assessed, or verified in production.
Read the published preprintEvidence model
Architecture statements should map to explicit invariants.
Implementation evidence should identify the tested commit and environment.
Negative tests and bypass tests are first-class evidence, not optional appendices.
Open implementation gaps, independent assurance, S6–S8 assurance, benchmark and penetration evidence must remain visible rather than implied away.
Research areas
Open questions
How should policy correctness itself be verified?
Which leakage channels remain outside the EEB boundary?
How can logging completeness be evidenced, not merely integrity?
What latency budget is acceptable for real-time authority resolution?
Which parts of the architecture are suitable for formal verification?
Public evidence policy
Publish what can be reviewed without publishing the control plane.
TenantSage may publish invariants, threat models, conformance methodology, synthetic examples and redacted test outcomes. Proprietary resolver logic, internal SQL, privileged topology, credentials and customer data remain outside the public evidence surface.
Collaboration
Academic, industry and standards review.
TenantSage welcomes research collaboration around governance verification, multi-tenant authorization, evidence lineage and alignment analysis against frameworks such as NIST guidance, OWASP LLM risks and applicable AI governance obligations. Alignment analysis is not a compliance claim.
Discuss collaboration