Database authority research
Fail-closed row-level security denies access when tenant, project or actor context is absent, malformed or unauthorised. For autonomous agents, this matters because workers are distributed and fallible: a missing session variable must produce no rows or an error, never broaden access across customer project state.
A permissive fallback turns a context bug into a cross-project data exposure. RLS policies should derive access from validated database context and reject unset values.
Application filters remain useful, but they are not a substitute for an authority check at the data boundary.
Happy-path tenant tests are insufficient. Verification should cover missing context, invalid identifiers, reused pooled connections and attempts to address another project.
Administrative and migration roles require deliberate bypass design and separate operational controls.
Nexus keeps durable multi-project execution state. Fail-closed database rules support separation between customer projects and between actors within an orchestration run.
RLS is one layer; repository credentials, worker isolation and application authorisation must align with it.
Nexus launch access
Register your interest now and we will let you know when the autonomous AI software engineering control plane is ready.
