One production AI system. One independent assurance package. $10,000.
A fixed-scope production assurance sprint designed to determine what an AI system can actually do, what it actually did, which boundaries it violated, what evidence proves the result, and what controls should change.
What the $10K engagement delivers
AI estate + authority map
Identify the target agent, model, tools, APIs, data paths, identities, permissions, owners, versions and dependencies.
Production trace intelligence
Correlate requests, model calls, context, tool calls, responses, actions, identities and policy decisions into reconstructable execution evidence.
Adversarial assurance
Exercise prompt injection, tool misuse, authority escalation, data egress, tenant isolation, MCP, memory and dependency failure paths.
Risk + policy decisioning
Convert observed behavior and findings into deterministic ALLOW, REVIEW or DENY decisions tied to explicit controls.
Runtime + human enforcement
Apply action-level authorization, approval requirements, denial, suspension and escalation at the execution boundary.
Evidence package
Produce provenance-linked findings, evidence records, integrity verification, assurance state and a machine-verifiable assurance passport.
Deployment gate
Check the exact system version against required assurance state before release or promotion.
Impact + reassessment
Map material model, prompt, tool, policy or dependency changes to affected controls and required re-evaluation.
Incident + remediation
Track detection, containment, investigation, remediation, retesting and verification rather than stopping at a finding.
$10,000 is the productized entry point.
The sprint covers one production AI workflow and its connected control surface. It is deliberately sold as a fixed-scope production engagement rather than an open-ended platform contract. Larger fleet deployments, private infrastructure, additional systems and ongoing operations can be scoped separately after the first engagement.