Observe and Prove. Then Enforce.
Miralto, Verity’s compliance evidence platform, is scoped deliberately to observe, understand, and prove — guidance only, never enforcement. Sentry AI Security is its sibling, built with the opposite charter: it’s the enforcement half. Where Miralto tells you what happened, Sentry AI Security can act on what’s happening right now.
A passive collector — Windows and Linux, no inbound ports required — watches for connections to AI-vendor domains and AI-API-key-shaped environment variables across the fleet. Every finding is sealed into a tamper-evident, hash-chained ledger and mapped to NIST AI RMF controls, surfaced on a dashboard as a governance-gap heatmap with an exportable vendor-egress report.
Six Capabilities, Live and Verified
An August 2026 build closed six capability gaps against a named competitor, Forcepoint — all shipped, merged to production, and independently re-verified line by line, not taken on a self-report. Sold and delivered as a managed Sentry offering, not a self-serve signup product.
Agent Data-Flow Visibility
Logs what an AI agent actually touched — not just whether its own claim of success passed.
Content-Block & Content-Redact Enforcement
Real-time DLP actions that auto-fire in production, opening an Arkon ticket on every finding — no human override, by design.
Cloud DSPM
Scans AWS S3 for exposed PII, credentials, financial, and health data. Observability-only — it flags, it doesn’t act.
Network-Level Shadow-AI Detection
Catches AI-tool usage via DNS and firewall logs — works even with no endpoint agent installed on the device.
UEBA Behavioral Risk Scoring
Flags sustained high-risk user behavior across identity-aware connectors — patterns, not single events.
Browser-Extension Capture Point
Blocks risky ChatGPT, Claude, and Gemini submissions at the browser. Still prototype-stage — not yet delivered as a shipped feature.
Security built into engineering, not bolted on after deployment — see Verity AI Engineering.
What the Export Actually Is
The exportable report is vendor-egress telemetry — a record of what connected to which AI vendor, when, and from where. It is not a true AI Bill of Materials, and we don’t call it one.
Content-block and content-redact enforcement has no human override. That’s a deliberate design decision, not a limitation we’re working around — a finding that can be waved through by whoever’s on shift isn’t enforcement.
Find Out What AI Tools Are Already Inside Your Environment
A scoped shadow-AI assessment, delivered by the Sentry team that operates this platform — not a self-serve trial.