Human responsibility
A person remains accountable for deciding whether and how an AI output is used.
Responsible and human-centred AI
Use AI in ways that people can trust.
Good AI use is not only about getting a working answer. It means protecting people, checking accuracy, respecting rights, explaining meaningful AI involvement, and keeping humans responsible for the outcome.
1. Six principles to remember
These principles apply whether you are writing a prompt, generating code, analysing information, or designing a system that affects other people.
Accuracy
Important claims, sources, calculations, and code must be checked before use.
Privacy
Collect and share only necessary information, protect it, and avoid exposing identifiable or confidential data.
Fairness
Look for unequal quality, stereotypes, exclusion, or harm across people and communities.
Transparency
Explain meaningful AI use, limitations, and decision processes to people affected by them.
Rights and safety
Respect consent, authorship, licences, cultural interests, security, and the wellbeing of others.
2. Risks to check before using AI
The level of review should match the possible harm. A private brainstorming task is different from a tool that influences grades, employment, health, finance, legal rights, public information, or access to services.
False or unsupported output
AI can invent facts, citations, cases, quotations, and explanations. Confident wording does not make an answer reliable.
Bias and exclusion
Training data and design choices can reproduce unfair patterns or work less well for particular languages, cultures, or groups.
Privacy and confidentiality
Prompts, uploaded files, logs, and outputs may contain personal or confidential information. Removing names alone may not prevent re-identification.
Copyright and provenance
Generated material can resemble protected work, and AI output may not provide reliable source history. Check permission, licensing, attribution, and originality.
Security and misuse
Generated code may contain vulnerabilities, unsafe dependencies, hidden network calls, or insecure handling of credentials and user data.
Over-reliance and automation
People can defer to a fluent system even when it is wrong. Consequential decisions need meaningful human review and a way to correct or challenge errors.
3. A responsible decision workflow
Use this process before sharing data with an AI service or acting on its output.
Define the purpose
What benefit is expected? Is AI necessary and proportionate, or would a simpler method work?
Identify who is affected
Consider users, non-users represented in the data, communities, creators, and people subject to decisions.
Classify the information
Remove unnecessary data. Do not enter sensitive, personal, confidential, or culturally restricted information without approved safeguards.
Test and verify
Check accuracy, fairness, accessibility, privacy, security, and performance with realistic and edge-case inputs.
Keep human oversight
Name the responsible reviewer, define when the system must stop or escalate, and allow errors to be corrected.
Explain and monitor
Disclose meaningful AI use, document limits and decisions, collect feedback, and review risks as tools change.
Stop and seek advice when:
The task could materially affect someone’s rights, safety, grades, employment, finances, health, legal position, privacy, reputation, or access to an opportunity—and appropriate approval or expert review is not in place.
4. AI in study and research
AUT guidance emphasises checking each assessment’s instructions, being open about permitted AI use, keeping submitted work genuinely yours, and asking for guidance when unsure.
Before starting
- Read the assessment instructions and course guidance.
- Confirm what AI assistance is permitted.
- Plan how you will record and acknowledge its use.
While working
- Do your own thinking and evaluate every output.
- Do not upload restricted research or participant data.
- Keep prompts, drafts, sources, and key decisions where appropriate.
Before submitting
- Verify facts and open every cited source.
- Check originality, attribution, and referencing requirements.
- Disclose AI assistance in the required form.
If uncertain
Ask your lecturer, supervisor, ethics committee, privacy officer, or other responsible authority. A general AI guide cannot override the rules for a specific course, research approval, workplace, or profession.
5. Final check before publishing, sharing, or acting
- The purpose is legitimate, clear, and suitable for AI assistance.
- No unnecessary personal, confidential, secret, or restricted information was entered.
- Important outputs were verified using authoritative evidence or appropriate testing.
- Bias, accessibility, cultural impact, and effects on different groups were considered.
- Copyright, licensing, attribution, consent, and academic-integrity requirements were checked.
- A named person has reviewed the result and remains responsible for the decision.
- Meaningful AI involvement and known limitations are disclosed where appropriate.
- People affected have a suitable way to report problems or challenge consequential errors.
Important: This page is practical educational guidance, not legal advice. Laws, policies, assessment rules, and product data controls can change; check the current requirements that apply to your situation.
Official guidance and further reading
Use current primary guidance when privacy, assessment, research, copyright, or organisational decisions are involved.
- AUT — Using AI in your studies
- AUT — AI in learning and teaching
- New Zealand Privacy Commissioner — AI and the Information Privacy Principles
- New Zealand Privacy Commissioner — Generative AI expectations
- NIST — AI Risk Management Framework
- Copyright Licensing New Zealand — AI and copyright fact sheet