As artificial intelligence becomes embedded in workplaces across New Zealand, most organisations are being told the same thing: use it safely, transparently, and in line with your existing values.
On the surface, this sounds straightforward. If you already have a code of ethics or a set of guiding principles, simply apply them to new AI tools and carry on.
But that assumption misses something important.
The hidden problem with “responsible AI”
Much of today’s responsible AI guidance treats values as if they are fixed — like rules written in a handbook. Once documented, the thinking goes, they can be translated into policies, checklists, and compliance frameworks.
Generative AI complicates that idea.
Unlike earlier technologies, these systems don’t just process information — they produce language. They write reports, draft policies, generate performance reviews, and produce the explanations organisations use to justify decisions.
And that changes something subtle but powerful: what people come to accept as a “good reason” for a decision.
How “value drift” begins
This is where “value drift” starts — a slow, often invisible shift in what an organisation considers normal, acceptable, or reasonable.
Because AI makes certain kinds of outputs faster, smoother, and more convincing, people naturally begin to rely on them. Over time, this can reshape everyday practices without any explicit decision to change values.
The drift often happens because the AI output feels helpful. But the cumulative effect is significant: explanations become more polished, decisions feel more neutral, and responsibility becomes harder to pinpoint.
Values are not fixed in practice
Traditional AI governance assumes organisations define their values first, then enforce them through systems and oversight.
But in reality, values are not just written down — they are lived.
They are shaped by daily routines: how people communicate, what they prioritise, and how decisions are justified under pressure. When AI becomes part of those routines, it subtly reshapes them.
Research in technology and ethics has long shown that tools and values influence each other. Technologies don’t simply reflect organisational values — they help form them.
We’ve already seen this pattern with social media. Practices that once felt intrusive or inappropriate gradually became normal, especially for younger users. The underlying value of privacy didn’t disappear, but its meaning shifted in practice.
Generative AI is likely to have similar effects on concepts such as fairness, accountability, and care.
Why the change is so easy to miss
Value drift is difficult to detect because it rarely appears as a clear failure or ethical breach. Instead, it unfolds through small, routine decisions that feel efficient and harmless at the time.
At the same time, organisations often lack clear visibility over how AI is shaping internal reasoning. Responsibility is fragmented between technical teams, legal teams, and business units — meaning no one is watching the full picture.
And because AI increases speed and productivity, it is easy for efficiency to outweigh reflection.
What organisations can do
Addressing value drift does not mean slowing AI adoption. It means strengthening how organisations observe and question its effects.
Rather than treating values as fixed commitments that simply need enforcement, organisations should treat them as something that must be actively maintained.
That could include:
- Regular reviews of AI-assisted decisions in sensitive areas such as hiring, performance management, or customer communication
- Auditing not just outputs, but the reasoning patterns AI introduces into decision-making
- Making clear who is accountable for decisions shaped or supported by AI
- Monitoring how concepts like fairness, transparency, or care may be subtly changing in practice over time
Responsible AI is not a one-off compliance exercise. It is an ongoing process of attention. The challenge is not only to ensure AI systems align with stated values today, but to recognise how those values may slowly evolve as AI becomes part of everyday organisational life.