Decision Desk · Issue 12 · complete public learning edition

Is It Time to Stop Using This AI?

Decide whether one exact AI use should continue, pause for repair, roll back, be replaced, or retire—and build the evidence, ownership, continuity, and exit record needed to act responsibly.

Begin the issue Open the decision record

Learning time90 minutes

Teaching12 cumulative lessons

InstrumentContinuation + retirement record

BoundaryNo automatic continuation

Four bounded dispositions

Continuation is one option—not the default.

The appropriate outcome depends on current evidence, current authority, accumulated effects, human control, and an executable exit—not on sunk cost, familiarity, or a past approval.

Continue with conditions

Only for the exact reviewed use, with current evidence and authority, named owners, monitoring, challenge, exit readiness, and a dated review.

Disposition

Pause and repair

Freeze expansion or use as required while resolving bounded evidence, control, access, configuration, or operational defects.

Disposition

Roll back or replace

Return to a known safer state or transition through a separately reviewed replacement plan with continuity and verification.

Disposition

Retire and archive

Stop the use, close access and integrations, preserve required records, complete data and vendor exit, support affected people, and verify residual obligations.

Disposition

Complete issue curriculum

Twelve cumulative lessons.

Use a fictional or nonsensitive system description. Never enter credentials, personal data, incident details, contracts, privileged material, confidential records, or security information.

Issue 12

Reopen the decision

Open

Treat deployment as a reversible decision, not a permanent achievement. Restate the exact use, purpose, people, owner, authoritative record, boundaries, approval conditions, and what has changed since the last review.

Issue 12

Test whether the purpose still exists

Open

Ask whether the original problem remains, whether the AI use still addresses it, whether a safer non-AI route now exists, and whether continuation serves affected people rather than merely preserving sunk cost or organizational habit.

Issue 12

Rebuild the current evidence record

Open

Use dated outcome, quality, error, override, complaint, incident, subgroup, workload, cost, and disconfirming evidence. A successful pilot, old benchmark, vendor claim, or absence of reported complaints does not prove present performance.

Issue 12

Recheck authority and affected-party conditions

Open

Verify current law, policy, contract, consent, notice, meaningful choice, access, challenge, records, privacy, security, accessibility, labor, sector, and decision-owner authority for the exact use. Prior permission may expire or stop fitting.

Issue 12

Freeze the system that actually exists

Open

Identify the current provider, model, version, prompts, tools, integrations, data flow, retention, training or reuse, subprocessors, permissions, human review, output path, and configuration. If the system cannot be frozen, record the uncertainty.

Issue 12

Look for accumulated harm and dependence

Open

Examine automation bias, deskilling, inaccessible alternatives, quiet scope expansion, workarounds, shadow systems, unequal burdens, suppressed dissent, downstream record errors, vendor lock-in, and whether people can still operate without the tool.

Issue 12

Separate repairable defects from stop signals

Open

A bounded defect may support controlled repair. Missing authority, severe or repeated harm, inability to inspect or correct consequential output, loss of meaningful human control, unsafe data handling, or no executable fallback may require pause or exit.

Issue 12

Design rollback before it is needed

Open

Define the last known safe state, manual fallback, trigger, stop owner, access changes, communications, records preservation, data export and deletion, continuity plan, verification, and maximum tolerable interruption. A rollback plan that has never been tested is an assumption.

Issue 12

Evaluate replacement without resetting the evidence burden

Open

A new vendor or model does not erase the decision history. Reassess purpose, authority, data, affected people, configuration, tests, human review, challenge, exit, cost, dependency, and transition risk before treating replacement as improvement.

Issue 12

Retire the use without erasing accountability

Open

Stop processing, remove access and integrations, preserve required decision and incident records, export or return data, request and verify deletion where appropriate, communicate the change, correct downstream records, support affected people, and document residual obligations.

Issue 12

Choose a bounded disposition

Open

Choose continuation with conditions, pause and repair, controlled rollback or replacement, or retirement. State the exact scope, owners, evidence, dissent, conditions, monitoring, stop rules, next review, and what the disposition does not authorize.

Issue 12

Keep the decision alive

Open

Monitor the conditions that justified the disposition. Reopen review after an incident, complaint pattern, material provider or model change, authority change, performance drift, new affected group, scope expansion, failed control, ownership change, or missed review date.

Critical review triggers

Past approval cannot carry the whole burden.

the current AI use, purpose, owner, affected people, authoritative record, or boundaries cannot be stated
the original justification no longer exists or a safer workable non-AI route is available
outcome, error, harm, complaint, override, subgroup, cost, or workload evidence is missing or stale
law, policy, contract, consent, notice, privacy, security, accessibility, labor, records, or sector authority is unresolved
the deployed provider, model, version, prompts, integrations, permissions, or data flow cannot be frozen
people cannot meaningfully challenge, correct, opt out where required, or reach accountable human support
the organization has lost the human capability or records needed to detect and correct failure
a rollback, continuity, data export or deletion, vendor exit, and verification path does not exist
serious or repeated harm, uncontrolled scope expansion, or consequential output without meaningful human control is present
the approval owner, stop owner, monitoring cadence, or next review date is absent

Interactive instrument · device-local

AI Continuation and Retirement Decision Record

Use a fictional or nonsensitive system description only. Do not enter credentials, personal data, confidential records, incident details, contracts, privileged material, or security information.

0/42continuation gates addressed

Past approval is not permanent permission.
Do not presume continuation. Freeze expansion and resolve missing or stale evidence.

Evidence area	Current evidence	Still authorized	Benefit persists	Harm controlled	Challenge works	Exit is executable
Purpose and accountable owner	Purpose and accountable owner: Current evidence	Purpose and accountable owner: Still authorized	Purpose and accountable owner: Benefit persists	Purpose and accountable owner: Harm controlled	Purpose and accountable owner: Challenge works	Purpose and accountable owner: Exit is executable
Outcome and quality evidence	Outcome and quality evidence: Current evidence	Outcome and quality evidence: Still authorized	Outcome and quality evidence: Benefit persists	Outcome and quality evidence: Harm controlled	Outcome and quality evidence: Challenge works	Outcome and quality evidence: Exit is executable
Harm, incident, and complaint record	Harm, incident, and complaint record: Current evidence	Harm, incident, and complaint record: Still authorized	Harm, incident, and complaint record: Benefit persists	Harm, incident, and complaint record: Harm controlled	Harm, incident, and complaint record: Challenge works	Harm, incident, and complaint record: Exit is executable
Authority, data, and affected-party conditions	Authority, data, and affected-party conditions: Current evidence	Authority, data, and affected-party conditions: Still authorized	Authority, data, and affected-party conditions: Benefit persists	Authority, data, and affected-party conditions: Harm controlled	Authority, data, and affected-party conditions: Challenge works	Authority, data, and affected-party conditions: Exit is executable
Exact provider, model, and configuration	Exact provider, model, and configuration: Current evidence	Exact provider, model, and configuration: Still authorized	Exact provider, model, and configuration: Benefit persists	Exact provider, model, and configuration: Harm controlled	Exact provider, model, and configuration: Challenge works	Exact provider, model, and configuration: Exit is executable
Human capability and operational dependence	Human capability and operational dependence: Current evidence	Human capability and operational dependence: Still authorized	Human capability and operational dependence: Benefit persists	Human capability and operational dependence: Harm controlled	Human capability and operational dependence: Challenge works	Human capability and operational dependence: Exit is executable
Rollback, replacement, and exit readiness	Rollback, replacement, and exit readiness: Current evidence	Rollback, replacement, and exit readiness: Still authorized	Rollback, replacement, and exit readiness: Benefit persists	Rollback, replacement, and exit readiness: Harm controlled	Rollback, replacement, and exit readiness: Challenge works	Rollback, replacement, and exit readiness: Exit is executable

Exact AI use, purpose, people, owner, and authoritative recordName the bounded use, intended outcome, affected people, accountable owner, human decision owner, authoritative record, exclusions, and what the system must never do.

Current outcome, quality, harm, incident, complaint, and override evidenceRecord dated measures, representative tests, subgroup effects, errors, incidents, complaints, challenges, overrides, workarounds, monitoring gaps, and disconfirming evidence.

Authority, data, provider, model, configuration, and dependency changesRecord changes in law or policy, consent or expectations, data flow, retention, training/reuse, provider terms, subprocessors, model, prompts, integrations, access, staffing, and operational dependence.

Rollback, replacement, data return/deletion, continuity, and archive planIdentify the last known safe state, manual fallback, replacement path, data export and deletion, vendor exit, records preservation, service continuity, verification steps, owners, dates, and irreversibilities.

Notice, challenge, support, monitoring, stop owner, and next reviewRecord affected-party notice, nonretaliatory challenge, correction, support, incident escalation, stop authority, approval owner, monitoring cadence, review trigger, and next dated review.

Bounded dispositionContinue this exact use with conditions, monitoring, and a dated reviewPause expansion and repair before any continuation decisionRoll back or replace the exact use under a controlled transitionRetire the exact use, preserve required records, and verify exit

Selection does not override the evidence floor.
A “continue” selection is only a proposed record. It is not approval and does not cure a concern, stale evidence, missing authority, unsafe dependence, or an untested exit.

Restoring your local record…

Completion boundary

The public course and reusable decision record are complete.

This edition helps structure a decision; it does not inspect a real system, validate evidence, provide professional advice, execute a rollback, delete vendor data, preserve required records, authorize continuation, or retire anything. Real action requires current domain-specific review, affected-party participation, accountable owners, tested continuity and exit procedures, and organizational approval.

Return to the 12-month program Previous issue