Rhindon Cyber logo
    Support & Documentation
    Contact Support

    AI Discovery & Catalog

    You cannot govern AI you do not know about. Discovery finds AI in use across your estate from browser and directory signals, matches it against a curated catalog of known AI services, and promotes confirmed findings into the AI Systems register.

    Where signals come from

    Browser extension
    The Shadow AI extension observes navigation and API traffic on managed browsers and reports which AI services were used, by which department, and how often. It does not capture prompt content.
    Microsoft Graph
    Directory scans surface AI-capable applications, OAuth grants, service principals, and non-human identities already consented to in your tenant.
    Microsoft Defender cloud discovery
    RAIC pulls in the AI apps your Microsoft Defender cloud discovery already sees in firewall and secure web gateway logs, so office and tunneled traffic counts too. When Defender for Endpoint feeds cloud discovery, that includes AI services reached by any app on enrolled Windows devices, such as desktop AI assistants, coding tools and scripts that call AI APIs.
    Microsoft agents and Agent 365
    Microsoft 365 discovery labels Microsoft AI agents (Copilot Studio, Foundry, Agent Builder and Agent 365) in the NHI registry. The optional Agent 365 registry connector lists every registered agent, and Defender alerts that name an AI agent open incidents.
    Agentless vendor collectors
    Read-only connections to Microsoft 365 Copilot, ChatGPT Enterprise and Claude (Enterprise/API) report usage from the vendor's own admin data. See AI Data Governance for setup.
    Manual entry
    Anything you already know about — vendor contracts, internal models, procurement records — entered directly.

    How detections are classified

    Every detection resolves to one of three entry types, because the governance questions differ for each.

    • Web — a person used an AI product through a browser. Detected from navigation to a known AI host. Governance focus: data leakage and acceptable use.
    • API — software called an AI service programmatically. Detected from XHR and fetch traffic to an API endpoint. Governance focus: integration inventory and vendor exposure.
    • Model — a specific model was invoked. Detected by parsing the model identifier out of the request path or body. Governance focus: model versioning, drift, and supply-chain provenance.

    The curated AI catalog

    The catalog is Rhindon's maintained reference set of known AI services, now past 1,500 entries across web, model, and API detections. Each entry carries a domain, application name, vendor, category, and enrichment fields describing data handling, training-data use, retention, and regional availability. The catalog is shared reference data — it contains no tenant information — and is versioned and published, so you can see exactly what changed between releases.

    Unknown signals

    A detection that does not match a catalog entry becomes an unknown signal. Unknown signals are triaged rather than ignored: each carries a confidence score, the evidence that produced it, and a proposed classification. Signals at or above the confidence threshold are promoted into catalog drafts automatically, and high-confidence patches to existing entries are applied automatically for a restricted set of safe fields. Everything else waits for review, where you can promote, merge into an existing entry, or dismiss it.

    • Click any signal to open the full record, including provenance, the proposed patch, and enrichment fields.
    • Bulk actions let you promote or dismiss a filtered set in one pass.
    • Merges fill only empty fields on the existing entry, so curated data is never overwritten.
    • Multi-tenant hosts are keyed by source URL as well as domain, so distinct tenants on a shared platform stay distinct rows.

    What discovery does not cover

    RAIC reports only what these sources can see. Plan around these limits:

    • RAIC has no endpoint agent of its own. AI use in desktop apps and scripts is visible only on Windows devices enrolled in your Microsoft Defender for Endpoint, through Defender cloud discovery.
    • Browsers without the extension, such as unmanaged or personal browsers, are not covered by browser discovery.
    • There are no connectors for Zscaler, Netskope or Okta. Their AI traffic shows up only if it also reaches Microsoft Defender cloud discovery.
    • RAIC does not read or scan prompts and responses, and it does not measure model drift or output quality itself.

    Promotion into governance

    Discovery ends where governance begins. Promoting a catalog entry creates an AI System record pre-filled with vendor, category, and enrichment data, ready for an owner, a lifecycle stage, and a risk tier. Usage volume and department breakdown carry across, which usually answers the first question a reviewer asks: how widely is this already being used?

    Privacy posture

    Discovery is designed to be defensible to your own workforce. The extension records which service was used and when, not what was typed into it. Signals are scoped to your organization and never shared across tenants. Deployment through Intune or ADMX makes coverage auditable, and the extension's permission set is documented on the Browser Extension page.