AI is entering the workplace faster than most organisations can properly discuss what that actually means.
It helps screen applicants, analyses employee data, recommends development opportunities, supports managers with feedback and increasingly influences decisions about people.
Much of the conversation still focuses on one question:
Does the technology work?
But I think we need to ask another one just as seriously:
What if the technology works exactly as intended, but the outcome still feels unfair, opaque or simply wrong?
That question sits at the heart of my research article From Principlism to Practice: A Five-Principle Safeguard Framework for Workplace AI, published in Springer Nature’s journal AI and Ethics.
The basic idea is quite simple.
Responsible AI at work cannot only be about efficiency, technical performance or compliance. It also has to be about people.
We already have enough AI principles
One thing quickly becomes clear when you look at AI ethics: we are not lacking principles.
Fairness. Transparency. Accountability. Human oversight. Responsibility.
Most organisations would probably agree with all of them.
The difficult part begins one step later.
What does fairness actually look like when an algorithm helps decide who gets invited to an interview?
What does human oversight mean if a manager simply accepts an AI recommendation because the system appears objective?
And how useful is transparency if an employee knows that AI was involved in a decision but has no realistic way to understand or challenge it?
This is the gap my research addresses.
Instead of adding another list of ethical values, I ask how five established principles can be translated into concrete safeguards for organisations.
Five questions organisations should ask about workplace AI
The framework builds around five principles: beneficence, non-maleficence, autonomy, justice and explicability.
They may sound academic at first. In practice, however, they lead to surprisingly simple questions.
1. Who actually benefits?
AI projects are often justified through efficiency.
Faster processes. Lower costs. Better data. More automation.
There is nothing wrong with that.
But if AI is introduced into people’s working lives, efficiency cannot be the only measure of success.
Does the system reduce unnecessary administrative work?
Does it improve access to learning or development?
Does it help people make better decisions?
Does it actually improve the quality of work?
The key question is therefore not simply whether AI creates value.
It is who benefits from that value and who carries the burden.
2. What could go wrong, even if the system works?
Some of the most important risks of workplace AI are not technical failures.
A monitoring system can function perfectly and still damage trust.
A recruitment tool can produce consistent results and still disadvantage certain groups.
A generative AI tool can write convincing performance feedback and still introduce inaccurate information.
That means organisations have to think beyond the question of whether a model is technically reliable.
They also need to consider psychological, relational, professional and organisational harm.
And ideally, they should do that before something goes wrong.
3. Can people actually challenge the AI?
“Human in the loop” has become one of those phrases that appears almost everywhere in discussions about responsible AI.
But a human being somewhere in the process does not automatically mean there is meaningful human oversight.
Imagine a manager receives an AI-generated recommendation.
The system looks sophisticated. The result seems objective. The manager is under time pressure.
How often will that recommendation genuinely be questioned?
Meaningful oversight requires more than an approval button.
People affected by important AI-supported decisions should be able to understand what happened, provide additional information and request a real human review.
And the person reviewing the decision needs the authority to reach a different conclusion.
Otherwise, human oversight risks becoming little more than a formality.
4. Is the system fair in practice, not just on paper?
When we talk about fairness in AI, the discussion often immediately turns to bias.
That matters. But workplace justice is broader.
- Who gets opportunities?
- Who carries new risks?
- Who can challenge a decision?
- Who understands the process?
- And who might be overlooked because their work, career path or circumstances do not fit neatly into the available data?
An AI system might improve overall results while making things worse for a smaller group of employees.
Looking only at averages can easily hide that.
Responsible AI therefore requires organisations to look at the people behind the numbers.
5. Who is responsible when something goes wrong?
Transparency is important.
But transparency without responsibility does not get us very far.
An employee does not need to understand every mathematical detail behind an AI model.
They do need to know what the system is being used for, what role it played in a decision, where its limitations are and who they can turn to if something appears wrong.
This is why the framework uses the principle of explicability.
It combines two things that belong together:
understanding and accountability.
Because explaining an AI system is only useful if someone is also responsible for what happens because of it.
AI ethics is not an IT project
This may be one of the most important points.
Workplace AI cannot be governed by IT teams alone.
It affects leadership, HR, organisational culture, employee participation, data protection and the way decisions are made.
The technology itself is only one part of the system.
That is why my paper proposes what I call Occupational AI Safeguard Rounds.
The idea is to regularly bring together the people who understand different parts of an AI system and its consequences.
That might include HR, IT, legal, data protection, leadership and employee representatives.
And instead of discussing only whether the technology performs well, they would ask questions such as:
Does this use of AI still make sense?
Who benefits from it?
What harms are emerging?
Do employees understand how the system affects them?
Can decisions actually be challenged?
And who takes responsibility when something goes wrong?
These are not questions that can be answered once during procurement and then forgotten.
AI systems change. Data changes. Organisations change. The way people use technology changes.
Responsible governance therefore has to be continuous.
The bigger question behind workplace AI
I do not believe responsible AI means stopping innovation.
Quite the opposite.
Innovation becomes more sustainable when organisations know what they are willing to use technology for and where they draw boundaries.
AI will increasingly influence how people are recruited, developed, evaluated and managed.
That makes AI ethics much more than a technical debate.
It becomes a question of leadership.
A question of organisational culture.
And ultimately, a question of what kind of workplace we want to create.
The most important question may therefore not be how intelligent our systems become.
It is whether the organisations around them become wise enough to use them responsibly.
Read the research
The full article is available here:
From Principlism to Practice: A Five-Principle Safeguard Framework for Workplace AI
Citation
Guttmann, M. (2026). From principlism to practice: a five-principle safeguard framework for workplace AI. AI Ethics, 6, 544.
https://doi.org/10.1007/s43681-026-01404-9