Responsible AI use changes with the consequences of the task.
Asking AI for ten headline ideas is different from using AI to recommend an action that affects safety, employment, money or another person's rights. The technology may be similar, but the required level of oversight is not.
A practical approach is therefore risk-based: as the possible consequences increase, so should verification, human review and control over the information being used.
Match the level of human oversight to the potential impact of getting the AI-assisted decision wrong.
Good AI use requires more than a good prompt.
Use AI with a clear purpose
Start with a defined task. AI should support a useful objective rather than being added to a process simply because the technology is available.
Protect information
Share only the information required for the task and follow applicable privacy, confidentiality, security and workplace rules.
Verify important output
Treat generated claims, calculations, recommendations and references as material to review—not automatically as verified facts.
Keep human judgement
People should remain responsible for decisions that require context, experience, accountability or consideration of consequences.
Respect rights and sources
Consider copyright, licences, attribution, permissions and originality when AI is used with existing creative or professional material.
Take responsibility for the result
Using AI does not transfer responsibility for what you publish, send, approve or act upon. The final user remains responsible for appropriate review.
Before trusting the output, consider what is at stake.
What happens if the AI output is wrong?
What information am I giving the AI system?
Which parts of the output require independent verification?
Could this output affect another person's rights, opportunities or treatment?
Am I using AI to support a decision or allowing it to make the decision?
Who is responsible for reviewing and approving the final result?
Drafting & brainstorming
Generating ideas, improving wording, creating generic checklists or producing first drafts that will be reviewed.
Normal human review may be sufficient, depending on the information involved.
Analysis & recommendations
Summarising operational information, analysing trends, comparing options or producing recommendations that may influence work.
Verify data, assumptions and conclusions before using the output.
Important decisions
Decisions involving safety, employment, legal rights, significant financial consequences, security or other high-impact outcomes.
AI should not replace appropriate qualified human judgement, controls or professional review.
Risk depends on context. The same AI output may be harmless as an internal draft but inappropriate if it is used directly to make an important decision without review.
Human review should add judgement—not just approval.
Clicking approve without examining the reasoning is not meaningful oversight. Human review should challenge the AI output, apply context and determine whether the result is appropriate for the real situation.
Define the task
Decide what problem AI should help solve and what should remain outside the AI workflow.
Control the input
Choose what information is appropriate to provide and remove unnecessary sensitive data.
Challenge the output
Look for unsupported assumptions, factual errors, missing context and overly confident conclusions.
Apply context
Bring workplace knowledge, professional experience, policy and real-world constraints into the decision.
Approve the action
A responsible person should decide whether the output is suitable to send, publish or act upon.
Learn from outcomes
Review mistakes and successful uses so prompts, controls and workflows improve over time.
The goal is not to manually redo everything AI produces. The goal is to place meaningful human judgement at the points where errors or poor decisions could matter.
AI identifies a possible cause of a warehouse stock variance.
After reviewing a limited set of transaction information, the AI suggests that an outbound processing error is the most likely cause.
Accept the explanation as the confirmed root cause.
The AI may have identified a plausible pattern, but a plausible explanation is not evidence that the cause has been proven.
Treat the suggestion as an investigation lead.
Review outbound records, receiving records, adjustments, movements, count accuracy and other relevant evidence before confirming the cause.
AI can accelerate investigation by helping identify what to check. It should not convert an unverified possibility into a confirmed operational fact.
Build responsibility into the process instead of adding it at the end.
Set the purpose
Be clear about the task, desired result and the role AI should play.
Consider the risk
Think about the information involved, possible errors and consequences of a poor output.
Control the information
Use only appropriate data and follow privacy, confidentiality and workplace requirements.
Use AI as an assistant
Give clear instructions, context and constraints while keeping the AI's role appropriately limited.
Check important claims
Verify facts, numbers, sources and assumptions according to the risk of the task.
Apply human judgement
Consider context and consequences before accepting the output or taking action.
Improve the system
Use outcomes and mistakes to improve future prompts, workflows and controls.
Use AI to strengthen human work—not to remove responsibility from it.
Responsible AI use means knowing what AI is doing, protecting the information involved, verifying important output and keeping meaningful human judgement around consequential decisions.
Know why AI is being used.
Control information and access.
Check what matters.
Keep human accountability.
Safe & Responsible Use
You have completed the four KRIYANO guides covering practical information protection, workplace AI, copyright awareness and responsible AI use.