Discovery, catalog, and violations

Security overview

The organization landing page summarizes posture, AI apps, sensitive-content events, high-risk applications, common detectors, and active coverage layers. Treat missing coverage as a deployment signal, not as zero risk.

AI app discovery

New organizations can open Discovery assessment to create a browser key, wait for the first signal, review the detected risk, and save a sanction decision. Monitor organizations can complete the read-only assessment; active enforcement remains a Protect feature.

Posture → AI Discovery & Governance correlates observations from browser, gateway, desktop, directory, network, and other configured sources. The summary answers what needs attention before presenting the full inventory.

For each application, review:

• sanction state: sanctioned, under review, shadow, or blocked;
• catalog risk score and factors;
• discovery sources;
• users, event volume, and last-seen time;
• the approved alternative shown to users when applicable.

Changing sanction state can change future policy behavior. Test the resulting policy bundle on a pilot sensor before using broad blocks.

Unified asset inventory

The same page also inventories observed applications, models, agents, MCP servers, MCP tools, and AI browser extensions. Models and agents are recorded when gateway callers identify them; MCP assets come from endpoint inventory and runtime tool enumeration. This is an organization-scoped operational inventory, not proof that every listed asset's content was inspected.

AI app catalog

The catalog provides vendor and data-handling facts used to explain risk. Review training behavior, retention, subprocessors, hosting region, and available compliance evidence before sanctioning a tool.

Catalog records support a security decision; they are not a substitute for the vendor contract, DPA, current trust-center evidence, or legal review.

Governance evidence is source-attributed and versioned. A changed fact creates a new evidence version and supersedes the older active value; an identical refresh is a no-op. Orthalon rejects private URLs, credentials, raw prompts, and unsupported fact fields from the evidence importer. Administrators should still review the source and effective date before using a fact in a legal or procurement decision.

Violations

The violation feed shows the user, AI app, policy decision, detector types and counts, time, and optional placeholder-only excerpt. It is designed for triage without becoming a prompt archive.

1. Filter by time, application, user, detector, or decision.
2. Open a record and review the policy that produced the decision.
3. Determine whether the event needs policy tuning, user coaching, provider review, or an incident workflow.
4. Export only when the current plan permits it and store the export according to your evidence-retention policy.

You cannot recover the original sensitive value from a normal Orthalon event. Ask the employee for context through your established security process instead of collecting protected content in Orthalon.