ascendraAI READINESS & ADVISORYBook Consultation

Illustrative engagement readout using fictional data. This sample is not a scored v0.3 assessment. Explore the current five-dimension framework →

ASCENDRA / SAMPLE READOUTFictional organization · Illustrative results

From AI visibility
to verifiable control.

See how a SAFE-AI-aligned assessment connects the AI inventory, control evidence and priority gaps to a secure AI control plane.

Northstar Group / Fictional sample
Illustrative evidence window: September 2026
SAMPLE / EXECUTIVE

A foundation to build on.
AI coverage to prove.

The sample organization has established security processes. The next step is to demonstrate that those processes cover its AI systems, data, identities and actions.

AI systems in scope06

Built, bought and embedded AI

Selected controls reviewed11 / 12

One control awaits assessment

Priority findings06

4 high · 2 medium

Improvement horizon90 days

Owners, dependencies and exit criteria

EXECUTIVE CONCLUSION

Keep the foundation.
Close the AI gaps.

Assessment and risk-management practices are evidenced in the sample. Approval, agent permissions, data boundaries and supplier controls need more complete AI coverage. Monitoring remains unverified.

Explore the target control plane
DECISIONS FOR LEADERSHIP
  1. Confirm the AI boundary. Resolve two pending uses and agree the discovery blind spots.
  2. Prioritize exposure. Constrain agent actions and protect sensitive data paths.
  3. Fund evidence of operation. Assign control owners and validate effectiveness before closing findings.
ASSESSMENT SNAPSHOT

Documented does not always mean demonstrated.

Each selected control is evaluated across requirements, operating evidence and explicit AI coverage.

Met2
Partial7
Not met1
No evidence1
Not assessed1

Illustrative judgments for this sample scope. “No evidence” means effectiveness is unknown; “not met” requires a confirmed gap. No aggregate compliance score or certification is implied.

AI CLARITY PROFILE / ASSESSMENT CONNECTION

Profile: not rated yet.

The Profile summarizes five AI Clarity dimensions. This fictional sample shows selected control findings, but does not contain evidence-backed ratings for all 25 criteria. No dimension scores are assigned.

Use the inventory and control findings as inputs, complete the Clarity rubric, and retain unknowns explicitly. The roadmap can still address the evidenced gaps while that work is completed.

Explore the five dimensions ↗ · Read the assessment method ↗

FOLLOW ONE FINDING END TO END

Knowledge agent → permissions gap → enforceable action boundary

Open the inventory record, inspect the control conclusion, and see the action and proof needed for closure.

AI-03 · Knowledge agentAC-6 · Least privilegeG-02 · Required improvement0–60 days · Validate and close ↗
SAMPLE / INVENTORY

Know the AI.
Know its boundaries.

A practical register connects each use case to its owner, data, model, identities, permissions and control findings.

Scope is explicit. Six known systems; four approved and two under review. Discovery reconciles sample tenant, browser, platform and supplier records. Personal devices and unmanaged accounts remain blind spots; estate completeness is not asserted.

Showing 6 of 6 sample systems

AI inventory — expand a system for scope and linked findings
AI systemTypeAccountable ownerApprovalRisk tier
AI-01Workplace assistantView inventory detail
Business purpose
Draft and summarize internal documents.
Provider / platform
Enterprise productivity suite
Model / version
Vendor-managed; version changes tracked by provider
Channel / hosting
Managed desktop and browser
Data boundary
Internal documents; confidential content only in approved locations
Identity & access
Employee SSO; source permissions inherited
Connectors & actions
Search and summarize; no write actions
Discovery / evidence
E-01 approval register + E-02 tenant export
Observation coverage
Managed tenant observed; unmanaged accounts outside this export.
Last sample review
30 September 2026
Linked findings
G-03 G-05
BoughtWorkplace technologyApprovedModerate
AI-02Research copilotView inventory detail
Business purpose
Summarize external research and draft market briefs.
Provider / platform
External research service
Model / version
Provider model and retention terms unverified
Channel / hosting
Public browser service
Data boundary
Public information intended; actual uploads not yet validated
Identity & access
Individual accounts; SSO coverage unverified
Connectors & actions
Web search and document upload
Discovery / evidence
E-03 managed-browser export + owner interview
Observation coverage
Observed usage; mobile and personal-device use not visible.
Last sample review
30 September 2026
Linked findings
G-01 G-03 G-04 G-05
BoughtResearch operationsUnder reviewHigh
AI-03Knowledge agentView inventory detail
Business purpose
Answer policy questions and create internal service tickets.
Provider / platform
Internal orchestration / hosted model endpoint
Model / version
Model alias recorded; exact release pending validation
Channel / hosting
Internal application and API
Data boundary
Internal knowledge base and restricted service records
Identity & access
Employee SSO + service identity; connector scope exceeds pilot need
Connectors & actions
Retrieve documents; create tickets; write actions held during review
Discovery / evidence
E-04 agent configuration + E-05 permission export
Observation coverage
Pilot configuration sampled; production access tests outstanding.
Last sample review
30 September 2026
Linked findings
G-01 G-02 G-03 G-04 G-05 G-06
BuiltAI platformUnder reviewHigh
AI-04Support assistantView inventory detail
Business purpose
Suggest responses for a human support representative.
Provider / platform
Customer-service SaaS
Model / version
Embedded vendor model; release notice process pending
Channel / hosting
Support console
Data boundary
Customer contact details and support conversations
Identity & access
Support SSO roles; vendor service access under review
Connectors & actions
Suggest only; representative approves every reply
Discovery / evidence
E-01 approval register + E-06 supplier pack
Observation coverage
Contract and configuration reviewed; provider-side telemetry limited.
Last sample review
30 September 2026
Linked findings
G-03 G-04 G-05 G-06
EmbeddedCustomer operationsApprovedHigh
AI-05Developer assistantView inventory detail
Business purpose
Assist coding and explain approved repository content.
Provider / platform
Enterprise coding service
Model / version
Vendor-managed model selection
Channel / hosting
Managed IDE and browser
Data boundary
Source code; secrets and restricted repositories excluded by policy
Identity & access
Engineering SSO; repository access groups
Connectors & actions
Suggest code; normal review and merge approval required
Discovery / evidence
E-02 entitlement export + E-07 development policy
Observation coverage
Licensed seats reconciled; personal extensions not fully observed.
Last sample review
30 September 2026
Linked findings
G-03 G-04 G-05
BoughtEngineeringApprovedModerate
AI-06Document classifierView inventory detail
Business purpose
Classify internal records for downstream human review.
Provider / platform
Internal ML platform
Model / version
Classifier v2.1; model registry entry retained
Channel / hosting
Batch pipeline
Data boundary
Internal business records and labeled training data
Identity & access
Dedicated workload identity; read-only input access
Connectors & actions
Write proposed labels; human review for sensitive records
Discovery / evidence
E-08 model register + E-09 validation record
Observation coverage
Registry and sample validation reviewed; change controls not assessed.
Last sample review
30 September 2026
Linked findings
G-05 G-06
BuiltData operationsApprovedModerate

Risk tiers are illustrative business-impact judgments. Approval does not mean every control is effective. Supplemental inventory discipline supports the assessment; it is not presented as a separate MITRE certification requirement.

SAMPLE / ASSESSMENT

From control intent
to evidence of operation.

Twelve selected NIST SP 800-53 Rev. 5 references from MITRE SAFE-AI Appendix D illustrate the assessment method. This is an excerpt, not a full assessment.

01 / REQUIREMENT

Is it defined?

Approved policies, standards, roles and criteria establish the expected control.

02 / OPERATION

Does it work?

Records, configurations, reviews and tests demonstrate operation for an agreed period and sample.

03 / AI COVERAGE

Does it cover this AI?

Confirm applicability across the environment, AI platform, models and data.

12 selected controls
Selected control results and evidence dimensions
Control / conclusionResultRequirementOperationAI coverage
AC-1
Policy and Procedures

AI access policy assigns owners, but service identities and exception reviews are not fully covered.

Sample evidence: E-07 policy; E-05 access review sample

Linked findings: G-01 G-02

PartialEvidencedLimitedPartial
AC-3
Access Enforcement

Role rules exist; allowed/denied retrieval tests for the knowledge agent have not been supplied.

Sample evidence: E-05 permission export; retrieval test request open

Linked findings: G-02

PartialEvidencedNot evidencedPartial
AC-4
Information Flow Enforcement

Data classifications are defined; AI destinations and blocked-transfer evidence are incomplete.

Sample evidence: E-07 data-use policy; E-04 connector configuration

Linked findings: G-03

PartialEvidencedLimitedPartial
AC-6
Least Privilege

The pilot agent connector has broader write permissions than its approved task requires.

Sample evidence: E-05 permission export and pilot scope

Linked findings: G-02

PartialEvidencedLimitedPartial
AC-20
Use of External Systems

External-service review is established; approval and monitoring of the research copilot remain unresolved.

Sample evidence: E-01 approval register; E-03 observed service usage

Linked findings: G-01 G-04

PartialEvidencedLimitedPartial
AU-2
Event Logging

Application logs are defined; actor, model, connector and policy-decision fields are inconsistent.

Sample evidence: E-10 sample log schema and retention standard

Linked findings: G-05

PartialEvidencedLimitedPartial
CA-2
Control Assessments

The scoped assessment program retains reviewed procedures, sample operating records and tracked actions for AI.

Sample evidence: E-09 assessment plan and signed review sample

Linked findings: No open finding for this scoped conclusion.

MetEvidencedEvidencedEvidenced
CM-3
Configuration Change Control

Model, prompt and connector change controls are scheduled for a follow-up review; no effectiveness conclusion is made.

Sample evidence: E-08 registry only; release evidence not reviewed

Linked findings: G-06

Not assessedNot reviewedNot reviewedNot reviewed
RA-3
Risk Assessment

Scoped use-case assessments retain threat scenarios, business impact, owners and treatment decisions. Open treatments remain tracked.

Sample evidence: E-09 use-case risk records and approval decisions

Linked findings: No open finding for this scoped conclusion.

MetEvidencedEvidencedEvidenced
SA-9
External System Services

Supplier obligations exist; data reuse, model changes and AI incident responsibilities need explicit confirmation.

Sample evidence: E-06 supplier assurance pack

Linked findings: G-04

PartialEvidencedLimitedPartial
SI-4
System Monitoring

No AI monitoring evidence was supplied within the review window. Monitoring effectiveness is unknown.

Sample evidence: E-11 monitoring request outstanding

Linked findings: G-05

No evidenceNot evidencedNot evidencedNot evidenced
SR-2
Supply Chain Risk Management Plan

The sample owner confirms that no maintained plan covers AI models, connectors and upstream dependencies.

Sample evidence: E-12 owner interview; plan absence confirmed

Linked findings: G-04

Not metAbsentNot evidencedAbsent
How to read the results and methodology

Met: the agreed criteria are evidenced for the scoped sample. Partial: some criteria or AI coverage remain incomplete. Not met: a control deficiency is confirmed. No evidence: effectiveness cannot be concluded. Not assessed: review has not been performed.

These labels and three evidence dimensions are Ascendra’s illustrative assessment presentation, not a MITRE rating scale. System applicability, criteria, sample period and residual risk must be agreed for a real engagement. All evidence IDs, organizations and judgments on this page are fictional.

Framework reference: MITRE SAFE-AI, April 2025 — Appendices D and E ↗. The target control plane and roadmap are Ascendra recommendations, not a MITRE-prescribed product architecture.

SAMPLE / GAPS

Every gap needs
a control, an owner and proof.

Priorities reflect the fictional scenario’s data exposure, action permissions and uncertainty. Evidence gaps remain distinct from confirmed control failures.

G-01 / Coverage gapHigh

Bring observed AI into the approval process

Two observed systems do not have a completed approval decision. The register does not yet reconcile every discovery source.

AI systemsAI-02 AI-03ControlsAC-1 AC-20
Required controls & closure evidence
Business risk
Unreviewed use can bypass permitted data, supplier and accountability decisions.
Target state
One owned registry connects discovery, risk classification, approval, conditions and retirement.
Required action
Reconcile tenant, browser, procurement and platform records. Approve, restrict or retire both uses; document blind spots.
Accountable owner
AI governance
Evidence to close
Reconciled six-system register; two signed decisions; source coverage and exception log.
Use existing capabilities first
Extend existing inventory and intake workflow; add discovery only where source coverage is missing.
G-02 / Control gap + evidence gapHigh

Constrain what agents can see and do

The agent has excess connector privileges, and retrieval authorization tests have not been evidenced.

AI systemsAI-03ControlsAC-1 AC-3 AC-6
Required controls & closure evidence
Business risk
A manipulated or mistaken agent could expose restricted content or perform an unauthorized action.
Target state
Every human and workload identity has task-specific permissions; retrieval and tool actions enforce authorization.
Required action
Reduce connector scopes, retain human approval for writes, review service identities and test denied as well as allowed paths.
Accountable owner
IAM + AI platform
Evidence to close
Approved permission matrix; reduced scopes; access-review record; retrieval and action test results.
Use existing capabilities first
Reuse identity, secrets and application authorization controls; add action-level enforcement at uncovered tool boundaries.
G-03 / Coverage gap + evidence gapHigh

Enforce data boundaries across AI channels

Classification rules exist, but approved AI destinations and prohibited-flow checks are incomplete.

AI systemsAI-01 AI-02 AI-03 AI-04 AI-05ControlsAC-4
Required controls & closure evidence
Business risk
Sensitive data may leave approved boundaries through prompts, uploads, retrieval or connected tools.
Target state
Classification and permitted-use rules follow data from source to model, connector, output and retention boundary.
Required action
Map flows and permitted destinations. Apply classification-aware controls and verify representative block/allow scenarios.
Accountable owner
Data security + privacy
Evidence to close
Approved data-flow map; provider retention decisions; blocked/allowed transfer tests; reviewed exceptions.
Use existing capabilities first
Extend existing data protection, browser, network and application controls; evaluate an AI gateway only for uncovered paths.
G-04 / Control gapMedium

Make supplier and model dependencies accountable

AI-specific service obligations are incomplete and an AI supply-chain plan is confirmed absent.

AI systemsAI-02 AI-03 AI-04 AI-05ControlsAC-20 SA-9 SR-2
Required controls & closure evidence
Business risk
Unclear data use, component changes and incident duties can leave risk outside the operating model.
Target state
Approved dependencies have accountable owners, explicit service conditions and review triggers.
Required action
Record providers, models and connectors. Resolve training/retention terms, locations, change notices and incident duties.
Accountable owner
Third-party risk + procurement
Evidence to close
Approved dependency plan; responsibility matrix; supplier decisions; contract or exception records.
Use existing capabilities first
Extend procurement and third-party risk processes before introducing a separate AI supplier tool.
G-05 / Evidence gapHigh

Prove that AI activity can be detected and contained

Logging fields vary and no AI monitoring evidence was received. This does not establish that monitoring is absent.

AI systemsAI-01 AI-02 AI-03 AI-04 AI-05 AI-06ControlsAU-2 SI-4
Required controls & closure evidence
Business risk
Suspicious usage or data movement may not be attributable, investigated or contained promptly.
Target state
Owned telemetry covers in-scope channels and supports alerts, investigation, revocation and response.
Required action
Confirm log sources and retention; link identities, systems and policy decisions; exercise alerts and credential or connector revocation.
Accountable owner
Security operations
Evidence to close
Coverage register; traceable event sample; alert investigation; containment exercise with owner and timing.
Use existing capabilities first
Use existing logging and incident tooling; add collection or detection where AI-specific context is missing.
G-06 / Assessment gapMedium

Validate security when the AI system changes

Model, prompt and connector change controls were not assessed. The residual uncertainty needs an owner and review date.

AI systemsAI-03 AI-04 AI-06ControlsCM-3
Required controls & closure evidence
Business risk
A material change may invalidate earlier approval, permission assumptions or test results.
Target state
Changes trigger proportionate review, regression checks, release approval and a tested rollback path.
Required action
Assess existing release controls, define AI change triggers and validate a representative model or connector release.
Accountable owner
Engineering + security assurance
Evidence to close
Reviewed change procedure; one approved release with test evidence; rollback record; residual-risk decision.
Use existing capabilities first
Extend existing delivery and assurance workflows; evaluate additional testing only after coverage is understood.
SAMPLE / TARGET

A secure AI control plane.
Defined by what it can enforce.

A coordinated set of governance, identity, data, runtime and assurance controls across the AI estate. Each capability has an owner and verifiable evidence.

THE TARGET OPERATING STATE

Know what is used. Decide what is allowed.
Enforce the decision. Prove it works.

Coverage spans the environment, AI platforms, models and data—and the connections between them.

01

Discover & decide

Register every in-scope AI use and its owner.

CURRENT SAMPLE STATE

Registry coverage and two approvals unresolved.

TARGET CONTROL

Owned inventory, approval gates, risk tier, exceptions and retirement.

AI governanceClose G-01 ↗
02

Authorize identities & actions

Limit who and what can act.

CURRENT SAMPLE STATE

Agent privileges and authorization evidence incomplete.

TARGET CONTROL

Human and workload identities, least privilege, retrieval checks and approval for sensitive actions.

IAM + AI platformClose G-02 ↗
03

Protect data & execution

Apply policy where AI is used.

CURRENT SAMPLE STATE

Data destinations and enforcement coverage incomplete.

TARGET CONTROL

Permitted models and destinations, data controls, tool boundaries and tested block/allow decisions.

Data securityClose G-03 ↗
04

Govern dependencies

Control what enters the AI estate.

CURRENT SAMPLE STATE

Provider terms and dependency plans incomplete.

TARGET CONTROL

Approved providers, model and connector records, responsibility boundaries and change obligations.

Third-party riskClose G-04 ↗
05

Observe & respond

Trace activity and contain incidents.

CURRENT SAMPLE STATE

Logging partial; monitoring not evidenced.

TARGET CONTROL

Correlated events, defined detections, response ownership and tested revocation or shutdown.

Security operationsClose G-05 ↗
06

Validate & improve

Keep approval tied to evidence.

CURRENT SAMPLE STATE

AI change controls not yet assessed.

TARGET CONTROL

Release checks, recurring control review, closure validation and residual-risk acceptance.

Security assuranceClose G-06 ↗
Build from the current stack. Connect the inventory, identity provider, data protection, application controls, monitoring and delivery workflows. Add capabilities where validated gaps remain. A gateway alone cannot cover every embedded, browser, model or agent path.

The ideal state is a measurable operating target. Closure requires agreed tests and evidence, with residual risk accepted by the accountable owner; no architecture eliminates all AI risk.

SAMPLE / ROADMAP

A route from findings
to a defensible target state.

Sequence the highest exposures first, then verify operating effectiveness. These time bands are a sample plan, subject to scope and dependencies.

01
0–30 days

Establish the boundary

Accountable lead: AI governance + IAM

  • Reconcile the inventory and resolve both approval decisions.
  • Reduce pilot-agent permissions; retain the write-action hold.
  • Assign owners to all open control and evidence gaps.
Dependency
Sponsor agrees scope, risk owners and evidence access.
Exit criteria
6/6 known records have owners and allowed-use conditions; 2/2 pending uses have decisions; excess agent scopes removed.
Findings addressedG-01 G-02
02
31–60 days

Close the control gaps

Accountable lead: Data security + third-party risk

  • Validate retrieval and tool authorization.
  • Implement the agreed data-flow and provider controls.
  • Approve supplier obligations and the AI dependency plan.
Dependency
Inventory, permissions and priority data paths are agreed.
Exit criteria
All agreed high-risk allow/deny scenarios pass or have approved exceptions; each scoped supplier has an accountable decision.
Findings addressedG-02 G-03 G-04
03
61–90 days

Demonstrate operation

Accountable lead: Security operations + assurance

  • Validate telemetry coverage and a containment exercise.
  • Assess a material AI change and verify rollback.
  • Reassess open findings and document residual-risk decisions.
Dependency
Enforcement is deployed and reviewable operating evidence is available.
Exit criteria
Each in-scope system has verified monitoring coverage or an owned exception; closure evidence is independently reviewed.
Findings addressedG-05 G-06
Connect the roadmap to the Profile. Record which AI Clarity criterion each initiative improves, then reassess with operating evidence. A planned action does not change a score. See how findings become recommendations ↗
CLOSURE GATE

Implemented → evidenced → independently reviewed → accepted

Update the inventory and control record, retain the validation result, assign residual risk, and set the next review trigger. A completed task alone does not close a finding.

YOUR AI CLARITY ROADMAP

What would this reveal
about your AI estate?

Start with the systems you use, the evidence you have and the decisions you need to make.

Discuss your assessment ↗

Public sample only. All system records, evidence references and results are fictional. No customer data or live integrations.