QData Identity Security

The control loop

Discover identities. Govern owners and lifecycle. Authorize the action. Protect only where a policy enforcement point sits on the path. Audit the decision. Shadow first.

What the product covers

Category pages for AI Agent Security and Non-Human Identities go deeper on those topics. Authorization and MCP Security are capabilities and surfaces — not a rename of this product.

• Inventory and governance

  Register agents and other non-human identities, discover what already exists, and assign an owner. Create, review, expire, and revoke. Without a connector, inventory is what you register yourself or report through an SDK — not a promise that every secret in the estate appears automatically.

• Authorization and policy

  Each check asks the same question: this subject, this action, this resource, this context. The answer is allow, deny, or require approval. Policy is something you can read, version, and simulate. Risk can raise or lower confidence; it is not the policy decision point.

• Runtime and audit

  A recorded decision is not a block. Protection exists only on a path that includes a policy enforcement point: SDK, security gateway, MCP gateway, sidecar, or connector. MCP tools are resources. Secrets belong in the same identity model. The trail is evidence of decisions — not a badge row.

One decision, three outcomes

The check is always the same shape: subject, action, resource, context. Risk can inform it. A language model does not make it.

• ALLOW

  The subject may perform the action on the resource in this context.

• DENY

  The subject may not. In shadow, this is recorded as a would-be deny. In protect, a PEP can block.

• REQUIRE_APPROVAL

  The action waits on a human. That is a decision outcome, not a notification after the fact.

Organization and Tenant

One Organization owns the people who sign in. Isolated Tenants own agents, NHI, policies, and resources. Policy and inventory do not leak across tenants.

A software house typically maps each customer environment to a tenant. An AI product team often starts with a single tenant and adds more when a second environment must not share agents or secrets.

Software houses

Delivery organizations that ship agents and automations into many customer environments. This is a job on this product page — not a separate marketing URL, and not a promise of white-label or custom domains.

• One organization, many customer environments

  The software house’s people live on the organization. Each customer environment is typically a tenant, so agents, NHI, policies, and resources do not mix.

• Agents shipped into someone else’s estate

  The identity that acts at the customer is still yours to inventory and authorize. Standing access that outlives the engagement is an orphan with a contract date.

• Shadow before you protect a client path

  Simulate would-be denies in the tenant before a PEP blocks a production tool call. That order is the same as everywhere else — the blast radius is just a customer’s.

Shadow, then protect

Simulate produces the same decision as protect, including would-be denies, without blocking the caller. Enable enforcement after a PEP is on the path and the trail has been reviewed. That order is the product, not an optional extra.

How protection actually attaches is on the security model.

What this is not

The first offer is identity security for agents and other NHI — not a ten-year IGA program and not a magic block.

• Not a full IGA suite

  Human identity governance at enterprise connector scale is not the first offer. This product starts with agents and other NHI.

• Not prompt DLP as the core

  Inspecting prompts is not the control loop. The loop is identity, policy, a decision, and enforcement where a PEP exists.

• Not a chatbot wrapper

  An agent is an identity that calls tools. Wrapping a model UI is not the same as authorizing an action on a resource.

• Not a block without a PEP

  Seeing an event, scoring risk, or opening a dashboard does not stop the caller. Attach a PEP before you claim protect.

Developer path

Register an agent, check authorization, see the decision. The snippet is conceptual until the public API host is live. Full DX copy lives on the developers hub.

authorization.check({
  subject:  "agent:invoice-agent",
  action:   "payment.execute",
  resource: "payment:9281",
  context:  { amount: 240, env: "prod" }
})

Related pages

• Authorization

  The check: subject, action, resource, context — including runtime.

  Open authorization

• MCP Security

  When the action is a tool call. Surface, not a second product.

  Open MCP Security

• AI Agent Security

  Agents as a type of NHI: tool calls, MCP, no human session.

  Open AI Agent Security

• Non-Human Identities

  Service accounts, workloads, machines, bots, APIs — the rest of the set.

  Open NHI

• Security model

  Observe is not enforce. Shadow before protect.

  See security

• Pricing

  Conversation frames and a quote. No invented list prices.

  See pricing

Questions