Governance & Human Intervention | UNAUTOMATED

The governing standard

The Ten Commitments work together. They are not a menu.

An organization cannot remove disclosure, accountability, privacy, fairness, or the right to human review and still claim the protection the full framework is designed to create. Responsible governance asks whether every commitment is present in the real process, not merely in a policy statement.

Especially central here

Commitment Two.
No artificial intelligence system may be used to deceive a person about whether they are dealing with a human being or a machine, when that distinction matters to the trust between them.
Commitment Four.
Every consequential decision made with the assistance of artificial intelligence remains the responsibility of a specific, identifiable person or institution, and no system may be offered as the final answer to why something happened.
Commitment Five.
Every person affected by a consequential automated decision retains the right to a human review, a clear explanation, and a real path to appeal, correct, or reverse that decision.
Commitment Six.
Fairness must be tested, not assumed, and any system that performs well on average must still be examined for how it fails the people the average conceals.

Before a system is deployed

The practices that make accountability visible.

Disclose the system's role

Use timely, understandable notice when an automated or AI-assisted system materially shapes a person-facing process. State the system's role, not only its name.

Keep a real human reachable

Give people a direct way to contact a responsible human when they need help, context, a correction, or a review of a consequential outcome.

Make review meaningful

Review is more than a signature. The reviewer needs relevant context, authority to change the outcome, and enough time to exercise judgment.

Monitor and correct

Track errors, complaints, overrides, and recurring points of confusion. Use what is learned to pause, improve, or retire a process when needed.

Practical governance guidance: adapt the depth of notice, review, documentation, and monitoring to the decision's impact on people. Applicable law varies by jurisdiction, sector, and use case. Seek qualified legal counsel for specific obligations.

Practical blueprint

A Human Intervention Pathway

A visible route that lets a person understand an automated process, ask a question, reach a responsible human, and receive a documented response or correction.

01
Notice before consequence
Tell people when an automated or AI-assisted system has a material role, what it does, and where to find more information.
02
Plain-language explanation
Explain the system's role, the information considered, the limits of the process, and the decision that remains humanly accountable.
03
Reachable human contact
Provide a real person, team, or accessible channel with a stated way to ask questions without navigating a dead end.
04
Meaningful human review
Give a qualified reviewer authority to consider context, correct records, depart from an automated recommendation, and explain the outcome.
05
Question, appeal, escalation
Offer a clear sequence for questions, review requests, corrections, and escalation, including a way to raise urgent or unresolved concerns.
06
Record, learn, correct
Keep an appropriate record of the system's role, the review, and the resolution so leaders can audit patterns and fix recurring problems.

A Human Intervention Pathway must be more than a customer-service link. It should give a person a usable route to a reviewer with the information and authority to consider their circumstances and correct the record or outcome when warranted.

AI use & human judgment

Helpful systems still require responsible people.

This governance material may discuss AI-assisted systems, but UNAUTOMATED does not decide how an organization should act. Use the framework to support accountable human review, verification, and context-sensitive decisions.

AI was used as a collaborator in the development and construction of this site. Certain content and interactive systems may also use AI-supported methods. That support does not replace accountable human review or decision-making.

AI-supported tools can be useful for information and exploration, but their outputs are not professional advice and should not substitute for human judgment, independent verification, or an informed review of the relevant context.

For legal, financial, medical, employment, benefits, insurance, safety, compliance, or other consequential matters, consult a qualified human professional—such as an attorney or an appropriate field specialist—before acting.

Visitors and organizations remain responsible for reviewing, verifying, and deciding how to use information produced or suggested by these systems. This is part of the Ten Commitments: keep human intervention, accountability, and the ability to question or correct a process visible.

Read the full AI use and human judgment disclosure

Where it matters most

Different contexts. The same obligation to stay accountable.

These are practical questions for leaders considering automated or AI-assisted systems. They do not make claims about how any employer, insurer, government, or public agency currently operates.

Employment and applicant screening

Can an applicant understand the system's role and ask a qualified person to reconsider relevant context or an apparent error?

Explain when software helps sort, rank, match, schedule, or otherwise support an applicant-facing process.
Name a reachable contact for accommodation requests, factual corrections, and questions about the process.
Require reviewers to examine job relevance, context, and exceptions instead of treating a score as the decision.

Public benefits and public-facing services

Can a resident understand what happened, submit missing or corrected information, and reach an accountable person before a harmful mistake becomes entrenched?

Use plain-language notices that distinguish an automated process from the institution responsible for the service.
Design accessible routes for questions, document correction, review, and escalation.
Keep records that allow staff to reconstruct the system's role and respond to patterns of error.

Insurance and coverage decisions

Can a person identify the basis of a decision, provide context, and obtain a responsible review instead of being left with an opaque outcome?

Clarify where automated analysis assists intake, triage, investigation, pricing, or service decisions.
Set a documented process for correcting information and requesting review by a qualified decision-maker.
Monitor complaint themes, overrides, and uneven outcomes to identify when a process needs attention.

Local government and municipal services

Can people with different access needs use a clear, practical route to question an automated process and reach the office responsible for resolving it?

Offer a public explanation of the purpose, limits, contact channel, and review route for high-impact systems.
Avoid making a digital-only pathway the sole way to seek help or correct a record.
Make vendor obligations and internal ownership visible before systems go live.

Foundations and grantmaking

Can applicants understand whether technology assisted screening or prioritization and receive a respectful, useful route for questions or factual corrections?

Tell applicants when automated tools materially support eligibility, sorting, or review workflows.
Preserve a way to surface missing context, accessibility needs, and errors in submitted information.
Review whether system design burdens smaller, less-resourced, or nontraditional applicants unfairly.

Adoption checklist

A working checklist for leaders.

01
Name the accountable executive, operational owner, and human review owner before launch.
02
Describe the system's purpose, affected people, inputs, outputs, limits, and decisions it can influence.
03
Write plain-language notice and test whether affected people can find and understand it.
04
Publish or provide a reachable contact and a Human Intervention Pathway for questions, correction, review, and escalation.
05
Define what a reviewer can see, what they can change, and when a system must be paused or overridden.
06
Set proportionate records for decisions, appeals, overrides, complaints, incidents, and corrective actions.
07
Monitor outcomes over time, including error patterns and the people who may be obscured by averages.
08
Review contracts, procurement terms, privacy practices, security, accessibility, and jurisdiction-specific legal obligations with appropriate counsel.

Accountable procurement

Questions leaders should ask vendors.

Procurement is a governance decision. A vendor's claim that a system is automated, intelligent, accurate, or industry-standard does not remove the organization's responsibility for what happens to people.

What decision or recommendation does the system produce, and what does it not determine?
What information does it use, retain, infer, or share, and how can an organization correct inaccurate records?
How can a human reviewer inspect the relevant context and override or reverse an output?
What documentation, logs, version history, and incident support will be available to the organization?
How is performance evaluated across different populations and real operating conditions, and what limits are known?
What changes can be made without notice, and what customer approval or review is available before material changes?
What support exists for accessible notices, questions, appeals, escalation, and recordkeeping?
What happens if the organization pauses, exits, or retires the system?

Sources and further reading

Start with primary materials. Then apply judgment to the context.

These public resources inform the questions on this page. They do not replace legal, technical, accessibility, procurement, or community expertise for a specific system.

When technology affects people, human judgment must remain reachable.