Skip to main content

Foundry Integration — W1

1. Purpose​

The Foundry integration supports W1 (Agentic AI Exposure Discovery & Assessment) by providing:

  • Visibility into autonomous execution surfaces within Foundry.
  • Deterministic execution identity context.
  • First-party execution evidence artifacts.
  • First-party configuration metadata relevant to outbound boundaries (if exposed).

The integration provides execution-plane and identity-plane artifacts only.

It does not evaluate exposure, compute reachability, assign risk, classify egress, detect drift, or perform scoring.

2. Execution Constructs in Scope​

W1 applies only to autonomous execution.

A Foundry construct qualifies when:

  • It can execute without an active human-initiated session.
  • It produces first-party execution records.
  • Execution records expose deterministic identifiers that reference the specific execution construct.

In scope:

  • AI-connected execution.
  • Agentic execution surfaces.
  • AI agent actions.

Excluded:

  • Human-initiated interactive sessions.
  • Constructs without first-party execution records.
  • Constructs whose execution records cannot be deterministically linked to a unique execution identifier.

If deterministic linkage cannot be established, execution status must be unproven.

3. Identity Context​

The integration must surface identity-plane artifacts sufficient to resolve:

Execution construct
→ Authentication configuration
→ Identity object

Identity binding is valid only when:

  • Deterministic identifiers resolve a uniquely identifiable identity object.
  • The identity object is observable via first-party metadata.

Execution identity may be:

  • Human (delegated or run-as authority recorded as a user principal).
  • Non-human (service or application identity).

If deterministic identity resolution is not possible, identity status must be unknown.

No heuristic identity inference is permitted.

4. Execution Evidence Signals​

Execution is considered proven only when:

  • A first-party execution record exists.
  • The record exposes a deterministic identifier.
  • That identifier references the specific execution construct without heuristic matching.

The integration must surface:

  • Execution records.
  • Deterministic join keys between execution record and execution construct.

If deterministic linkage cannot be established, execution status must be unproven.

5. Outbound / Integration Signals​

If Foundry exposes first-party metadata indicating outbound endpoints, the integration may capture:

  • Configured outbound endpoint host/base URL.
  • Integration endpoint references.

Constraints:

  • Endpoint-level only.
  • No payload or header inspection.
  • No inference beyond deterministically observable configuration.

If outbound configuration artifacts are not exposed by Foundry, no outbound artifacts are defined in W1 scope.

6. Ownership Signals​

If exposed by Foundry, the integration may capture:

  • Assigned owner identifiers.
  • Owner principal type.
  • Owner state (e.g., active / disabled), if deterministically observable.
  • Group ownership references.

If ownership metadata cannot be deterministically resolved, ownership must be surfaced as unknown.

No ownership inference is permitted.

7. Determinism Requirements​

The Foundry integration must adhere to the following:

  • First-party metadata only.
  • Deterministic joins only.
  • No heuristic or probabilistic correlation.
  • No expansion beyond deterministically observable boundaries.

If deterministic linkage cannot be established at any layer, output must explicitly indicate unknown or unproven.

The integration is a deterministic artifact provider. All exposure evaluation occurs in W1 logic.