Skip to content

Responsible AI resourceMenuHomeModel policyDecisionsResearch & approachGet support

Leadership guide/Decisions/Human judgment & automated action

Human judgment & automated action

Keep a person answerable.

Human oversight becomes meaningful when someone has the competence, time and authority to intervene.

Download this brief↓

HumanSkills recommendation · September 2026

In this guideRecommended positionYour local choicesA campus situationWho is responsibleLanguage to useEvidence of implementationSupporting sourcesDownload PDF→

Read, share and use with your team.

A considered starting position.

HumanSkills recommendationSeparate assistance, consequential recommendations and actions. Assign a responsible person, define what must be reviewed, and preserve a meaningful route to explanation, correction and human reconsideration. Authorize connected actions separately, within limited permissions.

A nominal human approval can become a rubber stamp if the reviewer cannot inspect the basis or correct the result. Faster decisions are not necessarily better decisions. Stakes rise when a system can change records, send messages or affect access to an opportunity.

What the evidence supportsPurdue limits sole reliance on AI for specified consequential decisions. North Shore includes human oversight and routes for questioning outcomes. NIST supplies a broader voluntary framework for managing AI risk.

• Name the accountable decision owner.
• Give reviewers the information and authority needed to challenge a result.
• Make correction, suspension and recovery possible.

What your institution decides.

The right arrangement depends on the purpose, consequences, applicable requirements and capacity of your institution. Use these options to make the choice explicit.

Assist

Consider this when: AI drafts, explains or organizes material for a person.

The tradeoff: Still requires verification appropriate to the task and protection of the information involved.

Recommend

Consider this when: AI influences a decision affecting a person or institution.

The tradeoff: Define the evidence a reviewer needs, when review occurs and how a person can challenge an outcome.

Act

Consider this when: AI can send, change, spend, approve or otherwise affect the world.

The tradeoff: Use explicit permissions, confirmation for consequential actions, action records, limits and a way to stop and recover.

Walk through the situation.

Illustrative campus caseAn advising assistant can change a schedule.

The same product can explain course requirements, recommend an individual schedule and change an enrollment record.

Those functions require different approval and accountability arrangements.

1. For explanations, establish current approved information and access to an advisor.
2. For recommendations, define review of academic and financial consequences and a correction route.
3. For record changes, explicitly approve the action, permissions, confirmation, audit record and recovery process.

What the team produces: An authorization that distinguishes advice from action and identifies the responsible owner.

Put responsibility in the right place.

Board & trustees

Ask who can explain, correct and stop consequential AI-influenced decisions.

Institutional leaders

Resource human review and ensure authority is matched to responsibility.

AI task force

Specify the boundaries between assistance, recommendations and actions for each use.

Apply these roles within your institution’s actual governance and delegated authority.

Language to build on.

Illustrative model clause

Consequential AI-supported decisions must have a designated human decision owner and a defined review and reconsideration route. Reviewers must be able to challenge the output and obtain appropriate evidence. AI systems may take actions only within documented permissions, with controls for confirmation, monitoring, suspension and recovery appropriate to the consequences.

Copy the clause

Review this clause with the full policy and local requirements. It is an implementation starting point, not a statement of measured consensus.

Model policy §6→Model policy §9→Model policy §11→

Know whether it is working.

• A named reviewer with relevant expertise and authority.
• Records showing substantive review of representative difficult cases.
• A usable human reconsideration route and tested stop mechanism.

When to reconsiderReassess when a system moves from advice to action, affects additional rights or services, or review becomes ineffective.

Read the supporting sources.

Selected precedents and guidance supporting this chapter. These sources do not imply institutional endorsement of HumanSkills or agreement with every recommendation.

• Purdue University: Artificial Intelligence Use ↗Institutional policy
• North Shore Community College: Artificial Intelligence Principles and Policies ↗Institutional policy
• National Institute of Standards and Technology: AI Risk Management Framework ↗Voluntary framework

Source review: 24 September 2026. Read the editorial approach.

Continue the workProtect the learning inside the work.

→

See the wider
evidence and choices.

Read the research comparisons and explore the annotated library across US colleges, universities and international institutions. Use the model language as a starting point for your own decision.

Keep people accountable.→Define authority when AI takes action.→Make the policy yours→

Put this decision into practice.

Open the operating guide and working record→