ContextIQ Agent Data PolicyEnterprise Preview

Turn existing data labels into enforceable agent-runtime controls.

ContextIQ resolves source classifications and user permissions, determines what representation an AI agent may receive, and produces a runtime-specific policy with evidence for every allow, restrict, or deny decision.

See an example decision

Enterprise data

Google DriveConfidential

ConfluencePlanned

ContextIQ decision layer

User entitlement
Data classification
Runtime capability

Resolve → Decide → Compile → Prove

Evidence attached to every result

Agent runtimes

NVIDIA OpenShellPreview path

Additional runtimes — planned

runtime-native policy

ALLOW · RAW
RESTRICT · METADATA
DENY

Enterprise outcomes

What a repeatable decision path gives your team.

Give AI platform teams a repeatable path through data-access review.
Reuse existing classifications instead of creating a separate agent taxonomy.
Produce runtime-native controls without maintaining one-off templates for every workload.
Preserve the evidence needed to review, reproduce, and audit each decision.

The gap

The employee can open the file. Should their agent?

Enterprise systems already know who can open a document and how it is classified. Agent runtimes usually do not. Security teams must manually translate source permissions, confidentiality rules, runtime capabilities, and task context before an agent can safely receive data.

User path

Alice → Confidential file

User may read

Agent path

Alice → Coding agent → Confidential file

Separate decision required

Task
Runtime
Network
Persistence
Model destination

Source permissions are not agent permission

A user’s ability to read a file does not automatically authorize an autonomous process to retrieve, retain, or transmit it.

Every runtime expresses controls differently

Filesystem scope, network egress, credentials, persistence, and isolation are configured differently across runtimes.

A generated policy still needs proof

Security reviewers need the source facts, policy version, generated controls, and final decision—not an unexplained score.

Product workflow

Four steps from source label to runtime control.

01
Resolve
Read the effective classification and current-user entitlement from the source system.
02
Decide
Apply the organization’s agent data-handling rules.
03
Compile
Translate required controls into the selected runtime’s native policy.
04
Prove
Return the decision, generated controls, limitations, and evidence.

Source entitlement ∩ enterprise data policy ∩ runtime capability = effective agent access.

Illustrative decision

One decision the runtime can enforce—and a reviewer can explain.

Illustrative response

{
  "decision": "allow_restricted",
  "subject": "alice@example.com",
  "resource": "drive:file-123",
  "classification": "confidential",
  "representation": "raw",
  "requiredControls": [
    "read_only",
    "isolated_session",
    "no_arbitrary_egress",
    "no_persistence"
  ],
  "runtime": "openshell",
  "policyVersion": "2026-10-21.1",
  "expiresIn": "30m"
}

Alice may read the source file. The company permits agents to use Confidential data only in an isolated, non-persistent environment without unrestricted egress. This OpenShell profile satisfies those requirements, so ContextIQ permits a time-limited, read-only copy.

Decision CIQ-2048 ALLOW · RESTRICTED

Resource: drive:file-123
Classification: Confidential
Subject: alice@example.com
Representation: Raw, read-only
Runtime: OpenShell
Expires: 30 minutes

Isolated session
Read-only input
Approved model endpoint only
No persistent storage

Policy 2026-10-21.1 · 6 evidence items

Data representations

Control what the agent receives—not only whether it connects.

Ordered by content exposure, most to least. These are outcomes, not process steps.

Raw
Original content when the source entitlement and runtime controls satisfy policy.
Derived
A deliberately transformed or redacted representation.
Metadata only
Identity, classification, schema, or document properties without content.
Denied
No release when authorization is missing, classification is unknown, or the runtime cannot enforce required controls.

FAQ

Questions security teams ask first.

Does ContextIQ replace DLP or DSPM?

No. DLP and DSPM products discover, classify, and monitor sensitive data. ContextIQ is intended to consume those signals and translate the resulting handling requirements into agent-runtime controls.

Does a user’s source permission automatically authorize their agent?

No. Source entitlement is necessary but not sufficient. ContextIQ also evaluates the organization’s agent policy, requested representation, task context, and runtime capabilities.

What happens if a runtime cannot enforce a required control?

The decision fails closed or permits a safer representation, depending on the administrator’s policy. ContextIQ should never silently omit a required control.

Which integrations are included in the preview?

The first validation path focuses on Google Drive classifications and current-user capabilities translated into NVIDIA OpenShell policy. Other source and runtime adapters remain planned until verified.

Is the Enterprise Preview generally available?

No. ContextIQ is seeking design partners with an existing data-classification program and a concrete agent workload to validate.

Bring one classified data source and one agent workload.

Work with ContextIQ to define the access decision, compile the runtime controls, and verify the result against a real enterprise workflow.

We’ll follow up to learn your source system, runtime, current classification scheme, and the agent task you want to approve.

Last reviewed: October 21, 2026. Integration and runtime status will change as adapters are verified. See all ContextIQ tools