Identity Exposure
Security failures come from relationships too.
Identity tools tell you who has access. They stop at the permission boundary — exactly where the interesting question starts. Iverson traces identities through roles, cloud permissions, and Kubernetes authorization into the infrastructure those permissions actually reach, then tells you which access change removes the most exposure without breaking production.
Four questions, answered with evidence
What can this identity actually reach?
Not what permissions it was granted — what those permissions ultimately touch. Iverson follows group memberships, role bindings, and application assignments all the way into the services and databases at the other end.
Why does it have that access?
Every answer carries its chain: this group, that binding, this role, that rule. Explainability is a feature of the engine, not a report we write afterward.
If it were compromised, what would be affected?
Reachable resources are joined to your live dependency graph, so exposure includes the downstream blast radius of everything that access touches — the same chaos-validated model that scores your infrastructure.
Which change removes the most exposure without breaking production?
Every candidate removal is simulated on a logical copy of the graph — never applied — and ranked by risk reduction against operational impact, with the evidence and its confidence shown.
What we refuse to do
Most of the value in security tooling is in what it declines to claim.
Deterministic, never guessed
No language model decides whether an identity has access. Authorization is resolved in code, per provider, and unrecognized permissions are reported as UNKNOWN rather than assigned a convenient class.
Absence of logs is not proof
"Not observed" only becomes "unused" when telemetry coverage supports it. Without coverage, the honest answer is insufficient evidence — and that is what you get.
Read-only, and never revokes
Connectors read. Nothing is revoked automatically. Remediation ends where it should: a reviewable change with the modeled impact attached.
Consequence, not probability
Identity blast radius says what a compromise could reach. It never pretends to know how likely a compromise is.