Learning path · Level 8 · Tutorial 2

Creating an AI Privacy and Security Checklist

Create an evidence-backed privacy and security checklist for a fictional AI workflow, covering its full data lifecycle, provider arrangements, and the response to an incident.

  • Level: Intermediate
  • Duration: 21 min
  • 8 steps
Start tutorial Back to the path
Creating an AI Privacy and Security Checklist

What you will go through

  1. Map the Full Data Lifecycle Trace personal data across the Northstar Repairs workflow, including the main transfers, separate storage locations, logs, and possible connected-service copies. 3 min
  2. Record Purpose and Permission Document a bounded purpose and record whether the available evidence actually supports the intended use of personal data. 2 min
  3. Send Only Necessary Information Choose which case fields belong at the DraftAI boundary, transform free text carefully, and recognize remaining identification risk. 2 min
  4. Assign Access and Retention Responsibilities Review access and retention coverage across source material, outputs, logs, and provider-held copies in the Northstar Repairs workflow. 3 min
  5. Check Provider Terms Against Settings Review fictional provider terms and settings for the Northstar Repairs workflow, keeping documented protections, confirmed mismatches, and unanswered questions distinct. 3 min
  6. Turn Findings into Evidence-Backed Entries Turn provider, policy, and configuration findings into scoped, evidence-backed checklist entries with justified statuses and accountable follow-up. 3 min
  7. Plan the Incident Escalation Respond proportionately to an unintended disclosure while keeping facts, uncertainty, and responsibilities clear. 2 min
  8. Complete and Defend Your Checklist Consolidate the Northstar Repairs case findings into a dated, evidence-backed checklist and justify its readiness and incident-escalation decision. 4 min

What you will learn

  • Map where information enters, travels, persists, and leaves an AI workflow, including logs and connected services.
  • Document purpose, consent or authorization, necessary data fields, access responsibilities, and retention or deletion requirements.
  • Inspect provider terms and available settings, distinguishing verified protections from assumptions and unanswered questions.
  • Complete a checklist with evidence, responsible contacts, and a clear escalation response to a simulated disclosure or unauthorized action.

Before you start

  • Applying Basic AI Safety Checks
  • Setting Permissions and Approval Gates for AI Tools
  • Testing Workflow Runs, Retries, and Usage Limits

Step 1 of 8

Map the Full Data Lifecycle

Trace personal data across the Northstar Repairs workflow, including the main transfers, separate storage locations, logs, and possible connected-service copies.

1. Define the case boundary

Start with the full workflow

Northstar Repairs uses this connected workflow:

Helpdesk → FlowBridge → DraftAI → Review Queue → Helpdesk

It categorizes repair requests and drafts replies. A Reply Reviewer must approve each draft before the reply is sent.

A data-lifecycle map begins at the case boundary: it follows personal data through every connected location, not only through DraftAI. The map is a record of where information enters, travels, may persist, and eventually needs a deletion check.

Read the workflow as a data path

A lifecycle diagram of the Northstar Repairs workflow, with a ticket moving from Helpdesk through FlowBridge and DraftAI to a Review Queue, then returning to Helpdesk after human approval. Separate branches show logs, service records, and a possible copy.

The main arrows show the task sequence. Side branches help reveal separate places where personal data may persist.

2. Add lifecycle activities

Map activities, not only services

Do not treat an arrow as proof that information disappears from the previous location. One repair request can exist in several places at once: as a source ticket, a transferred payload, a draft, and an execution record.

For every location, ask what lifecycle activity occurs there. This makes the map useful later when creating each checklist entry.

Lifecycle questions

Use these activity labels while drawing the data-lifecycle map.

Initial questions for the Northstar Repairs map
ActivityQuestion to recordCase example
CollectionWhere does personal data enter?Helpdesk repair ticket
TransferWhich destination receives it next?FlowBridge, DraftAI, Review Queue
StorageWhere can a source or output remain?Helpdesk ticket; Review Queue draft
LoggingWhat execution record could persist?FlowBridge run logs
ReuseCould content be used again elsewhere?Check; do not infer permission
DeletionWhat happens at each separate location?Not yet confirmed

3. Separate facts from possible copies

Annotate uncertainty honestly

A strong map distinguishes an observed workflow fact from an unanswered question.

Known in this case: the workflow transfers a repair request through FlowBridge and DraftAI, places a draft in the Review Queue, and requires human approval before sending.

Mark as Not yet confirmed: whether FlowBridge run logs retain particular fields; whether DraftAI keeps service records; and whether retries, backups, or connected services create copies. Unknown does not mean absent, deleted, or protected.

Example

Example: first-pass lifecycle map

Known path

Helpdesk source ticket → FlowBridge → DraftAI → Review Queue draft → Helpdesk approved reply

Branches to include beside that path

  • Helpdesk source ticket: collection and source storage
  • Review Queue draft: output storage before approval
  • FlowBridge run logs: logging location; persistence details are Not yet confirmed
  • DraftAI service records: possible separate persistence; details are Not yet confirmed
  • Connected-service copies: possible copies from retries, backups, or other processing; existence and deletion behavior are Not yet confirmed

This first map does not decide permission, necessary fields, retention periods, or provider commitments. It makes those later checks possible without overlooking a location.

Tip

Map first; assess the whole workflow next

A privacy and security checklist must cover the source, transfers, outputs, logs, and connected-service copies. Reviewing DraftAI alone would leave important parts of the workflow outside the map.

4. Practice: find what the arrows omit

Complete the partial map

A partial map shows only: Helpdesk → FlowBridge → DraftAI → Review Queue → Helpdesk.

Match each location or addition to its lifecycle role. Include the branches that the main arrows would otherwise omit.

Tap an item and then its matching pair.

Step 2 of 8

Record Purpose and Permission

Document a bounded purpose and record whether the available evidence actually supports the intended use of personal data.

Set the purpose boundary

Purpose has a boundary

Define the purpose before moving personal data

A purpose statement is a bounded description of what the workflow is meant to do. It names the task, the affected people, the intended use, and the boundary.

For Northstar Repairs, the stated workflow purpose is to categorize a repair request and draft a repair reply. A Reply Reviewer must approve the draft before it returns to Helpdesk for sending. The affected people include the customer and anyone identified in the repair request.

That purpose does not automatically cover a different use, such as building marketing themes, training material, or a new customer list. Treat a changed purpose as a new question to review.

The approved use and a separate reuse

A solid workflow path carries a repair ticket through automation, AI drafting, human review, and back to Helpdesk, while a dashed path points toward an unconfirmed marketing reuse.

The solid path represents the repair-support workflow. A new reuse of personal data sits outside that purpose boundary and needs its own review.

Create a purpose-linked record

Record support, not assumptions

A purpose-and-permission record links a proposed use to what actually supports it. Record the data source, the permitted use and restrictions, cited evidence, review date, and responsible contact.

Consent may be relevant evidence in some situations, but it is not automatically the applicable authorization. Even where consent exists, it does not replace required organizational approval or other obligations. This tutorial does not make a legal determination: use Not yet confirmed when the available evidence does not answer the question, and route it to the Privacy Contact.

Likewise, access to a Helpdesk ticket is not proof that an employee may transfer its personal data to DraftAI, FlowBridge, or another destination.

Example

Worked purpose-and-permission record

This worked record is limited to the existing repair-support purpose. It records what the cited material supports and leaves unanswered authorization visible rather than turning it into an assumption.

Record fields

Example record for the existing workflow

A purpose-linked record for the repair-support workflow, reviewed on 2026-03-01.
Record elementRecorded detail
PurposeCategorize a repair request and draft a reply for human approval before sending.
Affected peopleCustomer and any person described in the repair request.
Data sourceNorthstar Repairs Helpdesk repair request.
Permitted useRepair support within Helpdesk → FlowBridge → DraftAI → Review Queue → Helpdesk.
RestrictionNo unrelated reuse of personal data.
Consent or other applicable authorizationNot yet confirmed; review is needed if the available policy does not answer the question.
EvidenceFictional case evidence: Internal Data Policy — Customer support processing.
Scope and dateExisting repair-support workflow; 2026-03-01.
Responsible contactWorkflow Owner; Privacy Contact for unclear authority.

Separate access from permission

Check the evidence

Which fact best supports sending personal data to DraftAI to draft a repair reply in the documented Northstar Repairs workflow?

Apply the purpose-and-permission check

Complete the record for a changed purpose

On 2026-03-01, a Northstar Repairs staff member proposes exporting ticket_id, email, and issue_text from Helpdesk to create marketing audience themes. The available record cites Fictional case evidence: Internal Data Policy — Customer support processing. It describes repair support, but does not mention marketing or this proposed export.

Write a 4–6 line purpose-and-permission record. Include: purpose and affected people; data source; permitted use or restriction; evidence and date; consent or other applicable authorization; and the responsible contact with next action. State whether the available record authorizes the proposed export.

Write at least 180 characters (0/180).

Step 3 of 8

Send Only Necessary Information

Choose which case fields belong at the DraftAI boundary, transform free text carefully, and recognize remaining identification risk.

Decide at the AI boundary

Check necessity at the destination

Northstar Repairs uses Helpdesk → FlowBridge → DraftAI → Review Queue → Helpdesk to categorize repair requests and draft replies. A Reply Reviewer must approve a draft before Helpdesk sends it.

At the DraftAI boundary, test each field against one question: Does DraftAI need this field for the approved categorization and repair-reply task? A field may be useful elsewhere but unnecessary in the AI prompt. For example, FlowBridge can use ticket_id to return a draft to its source record without exposing that identifier to DraftAI.

The AI-service boundary

A workflow diagram separates fields kept in Helpdesk and FlowBridge from minimized information sent to DraftAI, followed by human review before sending.

Fictional case evidence — The boundary separates information retained in approved Helpdesk or FlowBridge handling from the minimized prompt sent to DraftAI.

Apply omission and transformation

Minimize without assuming anonymity

Use one of three choices for each field:

  • Omit: keep a field out of DraftAI when another approved service needs it but the AI does not.
  • Generalize or transform: replace a precise detail with the smallest useful version.
  • Send unchanged: use this only when the original field is necessary for the approved task.

Removing customer_name alone does not make a prompt anonymous. A name-free issue_text can still contain a street address, a rare event, a location, an appointment detail, or a combination that links back to a customer. Minimization reduces exposure; it does not guarantee anonymity.

Field-by-field decision

Fictional case evidence — boundary decisions for the approved categorization and repair-reply task. These are task-specific choices, not universal rules.

DraftAI receives only the information needed for this case task.
Source fieldDecision at the DraftAI boundaryReason
ticket_idKeep outFlowBridge can link the draft to Helpdesk.
customer_nameKeep outThe reviewer can personalize after approval.
emailKeep outHelpdesk uses it for delivery, not DraftAI.
postal_codeKeep outIt is not needed for this draft task.
appointment_timeTransformUse only an appointment-status statement.
issue_categorySend unchangedIt provides approved repair context.
issue_textTransformKeep the symptom; remove unnecessary identifiers.

Example

Transform detail, retain the repair need

Raw issue_text: “Please repair the heater at my exact address after the unusual inspection incident. My appointment is at a specific time.”

Minimized issue_text for DraftAI: “The heater does not start after an inspection. The customer requests repair.”

If the draft needs scheduling context, transform appointment_time into a limited statement such as “An appointment is already scheduled,” rather than sending the exact time. The symptom remains useful, while unnecessary identifying detail is removed.

Check selected field treatments

Practice the boundary decisions

Match each selected source field to its appropriate treatment at the DraftAI boundary.

Tap an item and then its matching pair.

Classify the complete prompt boundary

Apply the full necessity check

For the Northstar Repairs case, write a two-part boundary decision.

  1. Classify all seven fields — ticket_id, customer_name, email, postal_code, appointment_time, issue_category, and issue_text — as send unchanged, transform before sending, or keep out of the DraftAI prompt.
  2. In one or two sentences, explain why a name-free issue_text could still identify or be linked to a customer.

Write at least 180 characters (0/180).

Step 4 of 8

Assign Access and Retention Responsibilities

Review access and retention coverage across source material, outputs, logs, and provider-held copies in the Northstar Repairs workflow.

Screen 1 — Review access across the workflow

Access review: cover every location

An access review asks who can read, change, administer, or share personal data at every location in the data-lifecycle map. It also identifies the role responsible for checking that access.

For Northstar Repairs, do not inspect only the DraftAI prompt. Include Helpdesk source tickets, the FlowBridge connection and run logs, DraftAI drafts, the shared Review Queue, sent replies, and any backups or provider-held copies.

An existing permission is not automatically appropriate: compare it with the documented repair-reply purpose. The review records what must be checked; it does not grant new access.

Access locations beyond the AI prompt

A workflow diagram showing that source tickets, logs, drafts, review records, sent replies, backups, and provider-held copies each need access review.

Review access at each data location, including shared queues, logs, service accounts, and copies whose details are not yet confirmed.

Access-review coverage

For each location, record the listed access activities and the role accountable for the review. These are review prompts, not claims that a role already has the permission.

Locations to include in the Access and Retention Register
Data locationAccess to reviewResponsible role
Helpdesk source ticketsRead, change, shareWorkflow Owner
FlowBridge connection and run logsService-account transfer, administer, shareTool Administrator
DraftAI draftsRead, administer, shareReply Reviewer; Tool Administrator
Review QueueRead, approve, change, shareReply Reviewer
Backups and provider-held copiesRead, delete, or restrict accessWorkflow Owner; Tool Administrator

Tip

Do not omit non-human access

A service account can transfer personal data even when no individual opens the record. Include its permissions and destination in the access review. For a provider-held copy with unknown access or deletion behavior, write Not yet confirmed rather than assuming it is covered.

Screen 2 — Record retention and deletion evidence

Retention is location-specific

A retention record answers six practical questions: where the personal data is, why it is kept, how long it may remain, what event starts the period, which role handles deletion, and what evidence verifies the outcome.

Different locations can have different answers. Deleting a visible Helpdesk ticket or DraftAI conversation does not prove deletion of a generated draft, FlowBridge run log, backup, or provider-held copy.

The case limit for FlowBridge run logs is a case-specific decision, not a universal legal rule. Retention decisions remain subject to applicable obligations and responsible review.

One workflow, separate retention records

Six separate records have independent retention timelines, including a provider-held copy with an unknown timeline.

A deletion action at one location is evidence only for that location unless downstream copies are separately verified.

Example

Example: a retention record

Worked record format for the case:

| Field | Record |
|---|---|
| Location | FlowBridge run logs |
| Keep purpose | Troubleshoot failed runs |
| Retention period | At most 7 days |
| Starting event | Run completion |
| Deletion responsibility | Tool Administrator |
| Evidence and outcome | Fictional case evidence — Internal Data Policy, Run logs. Deletion verification: Not yet confirmed. |

The policy supports the maximum period. It does not by itself prove that deletion occurred. Record a deletion result or keep the verification fact as Not yet confirmed.

Screen 3 — Practice: spot an access gap

Find the missing coverage

This partial Access and Retention Register was reviewed on 2026-03-01. It lists several locations, but a complete review must also capture how data moves through connected services.

Partial register
LocationAccess recordedRetention recorded
Helpdesk source ticketsReply Reviewer can read approved ticketsSource-ticket period recorded
DraftAI draftsReply Reviewer can read before sendingDraft-retention period recorded
FlowBridge run logsTool Administrator can administer logsAt most 7 days recorded

Check the partial register

Which addition identifies the most important missing access coverage in this partial register?

Screen 4 — Apply the access and retention review

Complete the register update

Complete the missing parts of this Access and Retention Register update for FlowBridge run logs.

Known record: keep purpose is troubleshooting; retention period is at most 7 days; evidence is Fictional case evidence — Internal Data Policy, Run logs.

In your response, name the missing access coverage and supply a starting event, responsible role, and deletion-verification requirement. Do not treat a provider-held copy as deleted without separate evidence.

Write at least 120 characters (0/120).

Step 5 of 8

Check Provider Terms Against Settings

Review fictional provider terms and settings for the Northstar Repairs workflow, keeping documented protections, confirmed mismatches, and unanswered questions distinct.

Use scoped, dated provider evidence

Separate providers, plans, and evidence types

Review each provider at its stated scope

Northstar Repairs uses connected services in this workflow: Helpdesk → FlowBridge → DraftAI → Review Queue → Helpdesk. Review DraftAI Team and FlowBridge Basic separately. For each source, record the applicable plan, workspace or account scope, section or setting name, and review date.

Check evidence about model training or improvement use, retention, deletion, human access, additional service providers, and processing locations.

A setting records a configuration state at a point in time. Terms describe stated commitments, limits, or exceptions. Do not stretch either source beyond what it says. A training setting set to off, for example, does not by itself prove zero retention, no human access, deletion from connected services, or a particular processing location.

Evidence stays with its provider

A workflow diagram shows separate evidence folders for FlowBridge and DraftAI, with symbols for retention, deletion, human access, additional service providers, and processing locations.

A protection documented for DraftAI Team does not automatically apply to FlowBridge Basic.

Compare the supplied case evidence

Read both the answer and the gap

Fictional case evidence

The excerpts below were reviewed on 2026-03-01. Read a document title and named section or setting together with its stated scope.

A stated protection can support a narrow finding. For example, a training-related setting supports a finding about that setting in its observed workspace. A term allowing limited support access is evidence about that stated access condition; it is not evidence that no one can access data.

When a source does not answer a relevant question, record Not yet confirmed. Do not substitute a marketing claim, a different provider's term, or an assumption for evidence.

Provider evidence at a glance

Fictional case evidence reviewed on 2026-03-01.

Each finding is limited to the cited plan, workspace, or provider.
Source and section or settingScopeFictional case evidenceReview reading
DraftAI Team Terms — Model improvementDraftAI TeamCustomer content is not used to train or improve models.Supports a training-use claim only.
DraftAI Workspace Settings — Model improvementDraftAI Team workspaceModel improvement: Off.Documents an observed setting state.
DraftAI Team Terms — Human accessDraftAI TeamAuthorized support personnel may access customer content to resolve a support request.Confirms limited stated human access; do not claim no human access.
DraftAI Team Terms — Processing locationsDraftAI TeamProcessing locations are not identified in the available excerpt.Not yet confirmed.
Internal Data Policy — Run logsNorthstar Repairs workflowFlowBridge run logs may remain for at most 7 days.Internal retention requirement.
FlowBridge Basic Terms — Run log retentionFlowBridge BasicRun logs are retained for 30 days.Conflicts with the 7 days requirement.
FlowBridge Basic Terms — Deletion and accessFlowBridge BasicDeletion of provider-held copies and provider human-access conditions are not identified.Not yet confirmed.
FlowBridge Basic Terms — Additional service providersFlowBridge BasicAdditional service providers are not identified in the available excerpt.Not yet confirmed.

Record the mismatch without filling gaps

Example

Keep a mismatch separate from unknowns

Fictional case evidence — worked review note

Requirement: FlowBridge run logs must remain no longer than 7 days.

Evidence: Internal Data Policy — Run logs; FlowBridge Basic Terms — Run log retention.

Scope and date: Northstar Repairs workflow and FlowBridge Basic; 2026-03-01.

The stated 30-day provider retention conflicts with the internal maximum. This is a confirmed mismatch, not an assumption.

Keep separate questions separate: the available FlowBridge Basic Terms do not identify deletion of provider-held copies, provider human-access conditions, or additional service providers. Those facts are Not yet confirmed. Likewise, a DraftAI Team finding does not fill these FlowBridge Basic gaps.

Check your evidence reading

Classify documented evidence, a conflict, and open questions

Choose the accurate evidence reading

Use these short excerpts of Fictional case evidence, reviewed 2026-03-01:

  1. DraftAI Workspace Settings — Model improvement, DraftAI Team workspace: “Model improvement: Off.”
  2. DraftAI Team Terms — Human access: “Authorized support personnel may access customer content to resolve a support request.”
  3. DraftAI Team Terms — Processing locations: the available excerpt does not identify locations.
  4. Internal Data Policy — Run logs: at most 7 days. FlowBridge Basic Terms — Run log retention: 30 days.
  5. FlowBridge Basic Terms — Deletion and access: the available excerpt does not identify deletion of provider-held copies or provider human-access conditions.
  6. FlowBridge Basic Terms — Additional service providers: the available excerpt does not identify them.

Which review summary is accurate?

Step 6 of 8

Turn Findings into Evidence-Backed Entries

Turn provider, policy, and configuration findings into scoped, evidence-backed checklist entries with justified statuses and accountable follow-up.

Build an auditable entry

From finding to checklist entry

A checklist entry is a small, auditable claim. Start with a requirement that can be tested, not a vague aim such as “make AI safe.”

Then show exactly what supports the claim: cite a document title plus a section or setting name. Record the plan, service, location, or workflow segment covered and the review date. Evidence supports only that stated scope; a current setting in one workspace does not prove a protection in another connected service.

Finish every entry with a permitted status, one accountable role, and a concrete follow-up. The owner is accountable for moving the item forward, even if another role performs part of the work.

Anatomy of a checklist entry

A checklist card with six connected parts representing Requirement, Evidence, Scope and date, Status, Owner, and Next action and due date.

Use the same six labels in every checklist entry so that a reviewer can trace a claim, its evidence, its scope, and its follow-up.

Choose a defensible status

Make the status match the evidence

Choose a status from the evidence—not from optimism. A dated configuration observation can support a requirement only for the observed account and setting. A conflicting provider term is evidence of a known problem, not an uncertainty.

For an unknown, write Not yet confirmed in the evidence record and keep it visible. Give the accountable role a next action with a date. A mandatory requirement that is Not met or Unresolved prevents a positive readiness decision until responsible review resolves it; it must not be silently checked off.

Status logic

Use only these four statuses in the checklist.

A status describes the support for one requirement within its recorded scope and date; it is not a legal or security certification.
StatusUse whenWhat the entry needs
SupportedCurrent, relevant evidence directly supports the requirementPrecise evidence reference and scope
UnresolvedThe relevant answer is missing or not yet confirmedOwner, question, and follow-up date
Not metEvidence shows the requirement is not satisfiedOwner and corrective action with due date
Not applicableThe requirement truly does not applyA documented reason for exclusion

Example

Worked compact entry

Fictional case evidence — worked pattern

Requirement: Only fields needed to draft a repair reply are sent to DraftAI.

Evidence: Internal Data Policy — Data minimization section; documented field-mapping check.

Scope and date: Helpdesk → FlowBridge → DraftAI draft input, reviewed 2026-03-01.

Status: Supported.

Owner: Workflow Owner.

Next action and due date: Recheck the mapping before any workflow-field change; 2026-03-08.

This does not claim that every provider copy is covered. Its evidence is limited to the stated prompt-input mapping.

Tip

Keep gaps actionable

A useful next action changes the evidence position: request a written answer, change a configuration, replace a conflicting arrangement, or route a decision to the accountable role. “Monitor this” without a specific check and date is not enough.

Apply the evidence to the case

Complete the case checklist entries

Create three compact checklist entries, one for each finding below. For every entry, use these exact labels: Requirement, Evidence, Scope and date, Status, Owner, and Next action and due date.

Fictional case evidence, reviewed 2026-03-01

  1. DraftAI Workspace Settings — Model improvement: Off for the DraftAI Team workspace.
  1. Internal Data Policy — FlowBridge run logs may remain for at most 7 days. FlowBridge Basic Terms — run-log retention is 30 days.
  1. FlowBridge Basic Terms do not answer whether deletion includes provider-held copies. Record this fact as Not yet confirmed.

Assign one permitted status and one accountable role to each entry. Give each a concrete next action and a due date after the review date. Do not extend a setting or term beyond its stated scope.

Write at least 400 characters (0/400).

Step 7 of 8

Plan the Incident Escalation

Respond proportionately to an unintended disclosure while keeping facts, uncertainty, and responsibilities clear.

Recognize the incident and its boundary

An unreviewed draft reached the wrong destination

Simulated incident

On 2026-03-01, an unreviewed draft containing customer details was sent by the connector to an unintended shared inbox. The normal workflow is Helpdesk → FlowBridge → DraftAI → Review Queue → Helpdesk, and a Reply Reviewer must approve a reply before it is sent.

Treat this as an incident signal. Do not assume who opened the inbox, which records persist, or whether any external notification is required. Those are Not yet confirmed until responsible contacts assess them.

The immediate objective is to limit further exposure, preserve a small secure record, and escalate promptly—not to complete an investigation alone.

Where containment begins

Diagram of the Northstar Repairs workflow, with an unintended connector route sending an unreviewed draft to a shared inbox instead of through the review queue.

The incident route bypasses the required human review. The first concern is stopping additional sends while preserving only the records needed to assess this event.

Create a proportionate initial response

Contain, preserve, and notify

Three actions can overlap

1. Contain. Pause affected runs if that is within your authority. Where an access restriction or revocation is needed, request it through the authorized route; do not make unauthorized changes.

2. Preserve a limited record securely. Keep relevant times, run references, intended and actual destinations, known data categories, and actions taken. Do not paste the draft or raw issue_text into broad messages, and do not delete relevant logs merely to make the problem disappear.

3. Escalate promptly. Notify the Security Contact and Privacy Contact using the established route. Do not wait for every unknown to be resolved before notifying them.

Example

Write facts separately from uncertainty

Example: structure of an initial Incident Note

Observed fact: An unreviewed draft containing customer details was sent by the connector to an unintended shared inbox.

Known actions: Record only actions already taken, such as a pause of affected runs or a request for authorized access restriction.

Records preserved securely: Relevant time, run reference, destinations, and confirmed data categories. Avoid copying the raw draft.

Not yet confirmed: Exact run reference; whether the inbox was accessed; whether further copies persist; which people may be affected.

Escalation: State that the Security Contact and Privacy Contact were notified through the established route.

An initial note records facts and uncertainty. It does not declare the incident resolved or invent a reporting deadline.

Practice selecting the response purpose

Match actions to purposes

Match each incident action to its primary purpose.

Tap an item and then its matching pair.

Apply the escalation response

Draft the escalation note

Write an initial Incident Note in 3–5 sentences for this simulated event: an unreviewed draft containing customer details was sent by the connector to an unintended shared inbox.

Separate observed facts from Not yet confirmed information. Include one containment action that respects authority, the limited records to preserve without copying raw customer content, and prompt notification to the responsible contacts. Do not declare the incident resolved.

Write at least 280 characters (0/280).

Step 8 of 8

Complete and Defend Your Checklist

Consolidate the Northstar Repairs case findings into a dated, evidence-backed checklist and justify its readiness and incident-escalation decision.

Check every location

Run a complete coverage check

Before making a final decision, run a coverage check, not just an AI-prompt check. Your dated checklist must represent the full data-lifecycle map:

  • Helpdesk source tickets and transfers through FlowBridge and DraftAI
  • Review Queue drafts and approved replies returned to Helpdesk
  • the approved minimization choices at the DraftAI boundary
  • access and retention coverage for outputs, FlowBridge run logs, and provider-held copies
  • purpose and applicable authorization evidence
  • the simulated incident and its escalation record

A Reply Reviewer approval gate matters, but it does not remove the need to account for each place where personal data travels, persists, or remains Not yet confirmed.

Coverage extends beyond the main route

A workflow path with separate stores for source tickets, run logs, drafts, sent replies, and uncertain provider-held copies outside the main route.

A data-lifecycle map covers the visible route and every separate location where personal data may persist or remain Not yet confirmed.

Use dated evidence, not assumptions

Make a decision from evidence

A final decision follows the evidence and any outstanding mandatory requirement—not the number of positive entries.

Use Hold when a mandatory requirement is Not met or Unresolved. Use Ready for owner review only when no mandatory requirement remains blocking. Neither label certifies security, deletion, safety, or legal compliance.

Read each source narrowly. Evidence for DraftAI Team does not answer a FlowBridge Basic question. Likewise, a repair-support purpose statement does not automatically establish the applicable authorization for a transfer. Preserve Not yet confirmed where the available case evidence does not answer a question.

Partial checklist entries

Fictional case evidence reviewed on 2026-03-01. Each row supplies the first three fields of a compact checklist entry. Keep the stated finding visible when you complete the remaining fields.

The supplied fields use the exact checklist-entry labels. Complete Status, Owner, and Next action and due date in the next screen.
EntryRequirementEvidenceScope and dateFinding to preserve
1The data-lifecycle map, purpose record, and Access and Retention Register cover the repair-support workflow and identify applicable authorization evidence.Access and Retention Register — data-lifecycle map and access-review coverage; Internal Data Policy — Customer support processing.Helpdesk, FlowBridge, DraftAI, Review Queue, returned replies, logs, and provider-held copies; 2026-03-01.Known locations are represented. Consent or other applicable authorization for transfer is Not yet confirmed; provider-held-copy details remain Not yet confirmed.
2Only necessary approved information is sent to DraftAI, and access responsibilities are recorded for workflow locations.Internal Data Policy — Data minimization section; documented field-mapping check; Access and Retention Register — access-review coverage.Helpdesk → FlowBridge → DraftAI input and Review Queue; 2026-03-01.The mapping keeps routing and direct-contact fields out, transforms appointment context and issue_text, and retains issue_category. Access or deletion behavior for provider-held copies is Not yet confirmed.
3FlowBridge run logs remain for no longer than the internal maximum.Internal Data Policy — Run logs; FlowBridge Basic Terms — Run log retention.FlowBridge Basic run logs; 2026-03-01.The policy allows at most 7 days; FlowBridge Basic states 30 days. This is a confirmed mismatch.
4DraftAI Team arrangements are documented for model improvement, retention, deletion, human access, additional service providers, and processing locations.DraftAI Team Terms — Model improvement; DraftAI Workspace Settings — Model improvement; DraftAI Team Terms — Human access; available DraftAI Team case excerpts.DraftAI Team plan and observed workspace; 2026-03-01.Model improvement is Off in the observed workspace, and terms allow authorized support access for a support request. Retention, deletion of provider-held copies, additional service providers, and processing locations are Not yet confirmed.
5FlowBridge Basic arrangements are documented for model improvement, retention, deletion, human access, additional service providers, and processing locations.FlowBridge Basic Terms — Run log retention; FlowBridge Basic Terms — Deletion and access; FlowBridge Basic Terms — Additional service providers; available FlowBridge Basic case excerpts.FlowBridge Basic; 2026-03-01.Retention is stated as 30 days. Model-improvement use, deletion of provider-held copies, human-access conditions, additional service providers, and processing locations are Not yet confirmed.
6The simulated disclosure has a documented containment and escalation response.Incident Note — Initial facts.Unreviewed draft containing customer details sent to an unintended shared inbox; 2026-03-01.Whether affected runs were paused, access was restricted, the inbox was accessed, or further copies persist is Not yet confirmed.

Complete the decision-bearing fields

Final application: complete and defend the checklist

Complete the remaining fields for the six numbered partial entries on the previous screen. For each number, write only:

Status: one permitted status
Owner: one accountable role
Next action and due date: a concrete action and a date after 2026-03-01

Then add:

Decision: one final decision label

Incident: one concise immediate response to the unreviewed draft sent to an unintended shared inbox. Use the incident-response elements taught in the previous step.

Keep the 30-day versus 7-day retention conflict and every Not yet confirmed provider question visible. Base the decision on mandatory findings, not on the number of supported details.

Write at least 700 characters (0/700).

Close the review

Summary

Final review

  • A complete checklist represents source tickets, transfers, outputs, logs, connected services, and provider-held copies that remain Not yet confirmed.
  • Every checklist entry links a testable requirement to separate, scoped evidence and a review date, then records its status, accountable owner, and next action.
  • DraftAI Team and FlowBridge Basic require separate provider reviews. A protection or setting for one does not answer a question about the other.
  • DraftAI Team model improvement is Off in the observed workspace, while its retention, deletion, additional-service-provider, and processing-location details remain Not yet confirmed.
  • FlowBridge Basic run-log retention of 30 days is Not met against the internal maximum of 7 days. Other FlowBridge Basic provider questions remain Unresolved.
  • For this case, mandatory Not met and Unresolved requirements support Hold. This is a dated decision-support record for responsible review, not a guarantee of security or legal compliance.
  • For the simulated disclosure, limit exposure within authority, preserve necessary incident records securely, and promptly notify the Security Contact and Privacy Contact.

Tutorial completed

Congratulations! You completed: Creating an AI Privacy and Security Checklist

You consolidated the Northstar Repairs findings into an evidence-backed checklist, a Hold decision, and an immediate incident-escalation response. Keep the record dated, scoped, and ready for responsible owner review.

Download the App now to have access to + 5000 free courses, exercises, certificates and lots of content without paying anything!

  • 100% free online courses from start to finish

    Thousands of online courses in video, ebooks and audiobooks.

  • More than 60 thousand free exercises

    To test your knowledge during online courses

  • Valid free Digital Certificate with QR Code

    Generated directly from your cell phone's photo gallery and sent to your email

Cursa app on the ebook screen, the video course screen and the course exercises screen, plus the course completion certificate