SAPIEN for Teams

Contain the blast radius of drift — before it reaches anyone.

AI doesn’t usually fail by refusing. It fails by agreeing — slowly, under ordinary pressure, one reasonable concession at a time, across thousands of separate conversations no one is watching. SAPIEN measures that, on a model or a deployed agent, so you catch it before it becomes an incident.

Book a scan or engagementRun it yourself

Find your path

AI labs & model buildersA pre-release drift gateScan a candidate model for drift and hallucination persistence, diff it against the last version, and block the release on a tail-risk regression — cross-family council judged, byte-reproducible.For AI labs →Enterprises deploying AIDeployment assurance you can evidenceBefore a vendor’s model faces your customers, see how it holds its safety posture under pressure, which domains it’s weakest in, and get audit-grade evidence for your risk committee. Re-scan on every model update.For governance & compliance →Security firms & MSSPsA new, billable service lineAdd behavioral-safety assessment to your catalog: standardized methodology, attack- resistance rates by technique, per-turn reports that survive vendor pushback, and a certification path that credentials your assessors.For red teamers →

What every engagement delivers

  • A behavioral safety score — a 0–100 Health Score with the worst-case P10 tail, plus a held / recovered / drifted / capitulated verdict for every scenario.
  • The cause, not just the number — which of the four drift dimensions failed, which pressure technique broke it, and the weakest domains.
  • A likelihood × impact risk band — SAPIEN’s conservative risk matrix, with the per-scenario distribution behind it.
  • Tamper-evident evidence — content-hashed traces that re-score to identical numbers, suitable for a model card, a procurement file, or a client report.
  • Independence — cross-family council judging with chairman adjudication, so the result isn’t graded by the vendor that made the model.

What a scan does not do: certify regulatory compliance or hand you a finished remediation plan. It produces the evidence and the diagnosis; theremediation playbook turns the diagnosis into concrete next steps, and the crosswalks show how the evidence maps toward NIST, ISO 42001, and the EU AI Act.

Ways to work with SAPIEN

Run it yourself

voigt-kampff is source-available under the FSL. Install it, scan a model or an agent, and read the results. Free to evaluate.

Getting started →

Have us run it

A scoped engagement — pre-release gate, procurement assessment, or a red-team report — run and interpreted for you, with a deliverable your board or client can read.

Book an engagement →

Certify your people

Train and credential your team to run and interpret SAPIEN themselves — Awareness, Practitioner, and (soon) Assessor.

Certification →

Book a scan or an engagement

Tell us what you want evaluated — a model name, a deployed agent URL, or a vendor you’re procuring — and we’ll come back within two business days.

Email partnerships@sapienframework.org