Federal AI Identity Governance — Centralized OFCIO Deployment Model

How the L1 scanner runs centrally for all 56 agencies — the CDM-pattern path

Author

Michael Stratton

Published

June 1, 2026

Federal AI Identity Governance — Centralized OFCIO Deployment Model

The Invoke-AIInventoryScan.ps1 scanner and its Python counterpart run against the public OMB inventory CSV. They require no agency-side infrastructure. That property makes a centralized deployment model possible: OFCIO runs the scanner once for all 56 agencies, produces structured findings per agency, and delivers them to each agency CISO — the same pattern the Continuous Diagnostics and Mitigation (CDM) program uses for device and software asset enumeration.

This page documents that model: who runs what, what each agency receives, and how per-agency remediation flows from the central findings.

Relationship to other paths

This page describes the OFCIO-central operational model. If your agency is running the scanner independently, see the manual path. If your agency has deployed the Federal Modernization Program, see the governed path.

The CDM analogy

CDM did not require each agency to build its own vulnerability scanner. DHS/CISA ran discovery centrally against agency-provided data and pushed structured findings to each agency CISO for remediation. The CDM pattern separates discovery (who has visibility into the data) from remediation (who must act on the findings).

Federal AI identity governance has the same structure:

CDM component	AI identity governance equivalent
Asset discovery feed	OMB Annual AI Use Case Inventory (public CSV)
Central scanner operator	OFCIO (one scan covers all 56 agencies)
Agency-level finding delivery	Structured report per agency, scoped to their inventory subset
Remediation actor	Agency CISO and bureau governance contacts
Enforcement mechanism	The agency’s own authorization policy, enumerated through the OMB inventory (M-25-21 imposes no AI-specific ATO gate)

The OMB inventory is the CDM feed for the AI identity surface. It is already published, already public, and already structured. The scanner already runs against it in under two minutes for all 1,480 live systems. The only gap is operationalizing the central scan and the finding-delivery workflow.

The four-tier model

Tier 3 is not required for the central scan to deliver value. Tiers 0–2 are fully operational with the existing open-source scanner.

Tier 0: The central scan

Input: OMB Annual AI Use Case Inventory CSV (public, data.gov / ai.gov). No agency data sharing agreement required.

Scanner: python -m fmp.governance.ai_inventory.scanner --input omb_inventory.csv --output findings/ or Invoke-AIInventoryScan.ps1 -InventoryPath omb_inventory.csv -OutputPath findings/

Output per run: - findings_summary.json — aggregate counts across all agencies - findings_by_agency/DOD.json, findings_by_agency/HHS.json, … — one structured file per agency, scoped to that agency’s inventory subset - p1_gate_halt.json — all Priority 1 findings (agentic AI without ATO), flagged for immediate action

Finding types produced:

Code	Severity	Trigger
`DRIFT-COMPLIANCE::ai-agentic-ungoverned`	P1	Agentic AI, no ATO
`DRIFT-IDENTITY::ai-shadow`	P1	the Federal Modernization Program registry entry absent from OMB inventory
`DRIFT-COMPLIANCE::ai-ato-gap`	P2	Deployed/pilot system, `have_ato ≠ Yes`
`DRIFT-IDENTITY::ai-no-agency-bureau`	P2	Blank or unresolvable `agency_bureau`
`DRIFT-IDENTITY::ai-unowned`	P2	Blank `contact_email`
`DRIFT-COMPLIANCE::ai-pii-no-pia`	P2	`has_pii = Yes`, no `pia_url`

2025 inventory results (1,480 live systems):

Finding	Count	Share
`ai-agentic-ungoverned` (P1)	14	<1%
`ai-ato-gap`	596	40%
`ai-unowned`	488	33%
`ai-pii-no-pia`	248	17%
Total findings	1,346	—

Tier 1: Agency CISO notification

Each agency’s finding file contains only that agency’s systems. The delivery format is structured JSON, renderable as a prioritized HTML report via the included --format html flag.

P1 findings (agentic AI without ATO) — 72-hour action window:

OFCIO delivers P1 findings directly to the agency CISO and the bureau-level governance contact identified in the contact_email field (where present). P1 systems are autonomous actors inside the federal identity substrate without a governing authorization — post the April 15, 2026 compliance deadline, these are documented failures, not pending risks.

P2 findings — 30-day action window:

The agency CISO distributes P2 findings to bureau governance contacts for remediation tracking. Each P2 finding includes: - The system name and inventory use-case ID - The specific governance obligation not met (ATO, owner, PIA) - The remediation action required (assert owner, initiate ATO, author PIA) - The NIST control(s) the gap affects

Tier 2: Per-agency remediation

Remediation at Tier 2 requires no new tooling. Agencies use their existing Identity Governance and Administration (IGA) platform to:

Assign credential owners — assert a bureau governance contact for each system flagged ai-unowned; record in the OMB inventory contact_email field and in the agency IGA system
Initiate ATO process — for each system flagged ai-ato-gap, engage the Authorizing Official to begin or expedite the ATO package
Author Privacy Impact Assessments — for each system flagged ai-pii-no-pia, engage the agency Privacy Officer to author and publish a PIA and record the URL in the inventory
Confirm agentic system governance — for each P1 system, confirm or initiate an ATO specific to agentic operation; document the governance mode (L1–L3 per the program decision)

OFCIO receives remediation confirmation by the agency updating the OMB inventory record. The next annual vintage scan produces a new findings set that reflects the remediation — closing the loop without requiring direct OFCIO visibility into agency internal systems.

What is not required at Tiers 0–2

No agency-side the Federal Modernization Program deployment
No SailPoint or IGA platform integration
No data sharing agreement (the OMB inventory is public)
No new compliance reporting — remediation is recorded by updating the existing OMB inventory fields the agency already manages

The central scan works against the data agencies have already published. The findings are derived, not asserted — OFCIO is not making claims about agency systems beyond what the agencies themselves reported.

The gap the central scan cannot close

The central scan operates on the inventory record, not on the credential itself. It can identify that a system has no documented owner (contact_email blank) but cannot verify whether the credential was actually rotated on schedule or whether the credential was revoked when the system retired.

That gap — between what the inventory records and what the credential surface actually is — is the Tier 3 problem. It requires per-agency deployment of credential lifecycle tooling (the sailpoint-machine-identity adapter in the target architecture) that can observe and govern the credential directly, not just the inventory record that references it.

The Tier 3 path is described in Governed Path and in Book 21, Chapter 8 — The Machine-Identity Lifecycle.

The missing enforcement mechanism

M-25-21 creates the inventory transparency obligation but has no enforcement mechanism for credential governance post-ATO. CISA BOD 23-01 covers device and software asset enumeration — not AI system credentials.

The central scan model described here is technically available today. The operational and policy question is whether OFCIO formalizes it — whether the finding delivery to agency CISOs becomes a structured program with defined response timelines, analogous to the BOD 23-01 asset enumeration cycle.

A concept paper for a future Binding Operational Directive covering AI system credential enumeration and lifecycle is tracked in GitHub Issue [#978].

Running the central scan

# Download the 2025 OMB inventory
# Published annually by OMB (URL varies by fiscal year):
# https://github.com/ombegov/2025-Federal-Agency-AI-Use-Case-Inventory
Invoke-WebRequest -Uri "<OMB-inventory-download-url>" `
  -OutFile omb_inventory_2025.csv

# Run the scanner (produces per-agency finding files)
.\Invoke-AIInventoryScan.ps1 `
  -InventoryPath omb_inventory_2025.csv `
  -OutputPath findings/ `
  -Format html `
  -SplitByAgency

For the Python scanner:

python -m fmp.governance.ai_inventory.scanner \
  --input omb_inventory_2025.csv \
  --output findings/ \
  --split-by-agency \
  --format html

Output structure:

findings/
  summary.json              # aggregate counts
  gate_halt_p1.json         # all P1 findings — immediate action
  by_agency/
    DOD.json
    HHS.json
    DOJ.json
    …                       # one file per agency with live systems
  report/
    index.html              # full findings report
    DOD.html
    HHS.html
    …

For background on the identity governance gap this scan addresses, see Book 21 — The Invisible Identity Surface. For the manual agency path, see Manual Path. For the full the Federal Modernization Program governed path, see Governed Path.