Powerful AI, kept on a short leash

Grounding over guessing

Every attack-path step the platform reports is tied to evidence it actually observed — a finding, an exposure, a relationship in the graph. A guardrail drops any hypothesis that can’t cite its evidence before it ever reaches you, and the number of dropped paths is shown so you can audit the system’s restraint.

Humans approve consequential action

Discovery and analysis are read-only and non-intrusive. Anything that would actively touch your systems — validation, simulated reachability — is gated behind explicit human approval and bounded by a global kill switch that halts all active work organization-wide, instantly.

We don’t train on your data

Your attack surface, findings, and reports are yours. We do not use customer data to train foundation models, and we don’t share it across tenants. Reasoning models are used as engines for analysis, not as sponges for your secrets.

Transparency you can inspect

The reasoner is pluggable, its provenance is recorded on every report, and the deterministic substrate runs even with no LLM configured. You can always see which model produced a narrative — and trace the evidence beneath every claim.

Bounded autonomy

Autonomy is a dial, not a switch. The platform is designed to escalate capability phase by phase — from analysis, to gated validation, to broader automation — with governance and human oversight scaling alongside it at every step.