Governance is NOT the boring part of AI
Written by Eija-Leena Koponen, co-founder & CAIO.
Early June, Tivi (a Finnish tech-focused media) published their second annual list of Finland's 50 most influential women in IT. My name was on it with the description "Eija-Leena's broad perspective on responsible and sustainable use of AI is an extremely important angle in the current market."
I was kind of surprised, first to see my own name there, second how accurate that description felt right now, after a spring spent as an interim AI Governance Specialist at a client. On Monday morning the client’s team greeted me with flowers. That made the recognition land twice, in the best possible way.
What interests me here is why "responsible and sustainable use of AI" is suddenly so relevant, worth lifting up. A year ago, for example, that framing felt like you were a backward-looking Luddite, as if caring about governance meant you weren't really a believer of great progress. Today it reads differently.
"Responsible and sustainable use of AI" is suddenly relevant.”
Earlier this month I also gave a keynote at a company's summer day. The title was "Tired of AI or still waiting for the apocalypse?" The room was full of people who had spent two years being told AI would change everything, and were now quietly wondering what exactly had changed.
My answer was that even though we haven’t seen an apocalypse, quite a lot did change, but just not in the way the headlines and the Silicon Valley tech bros suggested.
Firstly, we use AI “badly”. Not because the tools were bad, but because the tools were designed carelessly looking for the first and shiniest output to push through, not taking into account societal or other things. Algorithmic fatigue has been real since Netflix scrolling. Spring 2026 has also brought the terms “AI brain fry” and “AI workshop” to our vocabulary. When an AI system generates every recommendation, every summary and email, every decision prompt, users stop trusting AI or themselves as experts, and then either stop using AI or go zombie with it. The problem is not adoption. It is the design (or lack of it).
Secondly, efficiency gains in AI did not reduce the overall demand for work. They expanded the surface area. More customers served, more channels opened, more content generated. Speed amplifies what is already there. A lot of AI activity produced marginal gains because it was deployed on the same processes, at the same scale, with the same assumptions as before.
Thirdly, we have built bias in, not out. Bias in AI systems does not usually arrive through malice, it arrives through shortcuts. The systems that were deployed fastest, at the highest volume, were often the ones with the least scrutiny. Which is basically all of the current new and shiny things. The most common shortcut is treating historical data as ground truth, testing on the people who built it and optimising for aggregate metrics. Let me remind you that women make up 22% of the global AI workforce and hold less than 14% of senior AI leadership roles. The system's default is always someone's normal (men), and someone else's exception (women and other minorities).
So no wonder that a study published earlier this year by the Oxford Internet Institute ("Women Worry, Men Adopt: How Gendered Perceptions Shape the Use of Generative AI", Stephany & Duszynski, 2026) surveyed around 8,000 people about their relationship with generative AI and found that women have lower adoption rates with AI and the fact has almost nothing to do with skills or access. Women who are highly educated, digitally confident, and entirely capable of using tools like ChatGPT remain more hesitant than their male counterparts. Their hesitation is driven by a different concern: the wider consequences. Employment effects. Privacy. Misinformation. Mental health. The environment. Women are not afraid of the tools. They are skeptical of what the tools represent, how they are built and how they are being deployed.
That skepticism is not a skills gap. It is informed judgment about systems whose design choices are still being made and/or are made “normal” in mind.
What has Governance to do with it?
Governance is a sequence of decisions that happen before the system reaches users. What data goes in, and who decided it was representative? What counts as an acceptable error rate, and acceptable for whom? Are the privacy, security and other contracts and/or settings in place? Who reviews outputs before they become actions? What happens when it fails? These are not only technical questions, they are also design questions. And the answers reflect whoever was in the room when they were asked, and whether anyone in that room had the standing to slow things down.
Regulation has been trying to catch up to the mess. The EU AI Act, which began applying in earnest in 2024, is the most serious structural attempt so far to address exactly the problems described above. It works on a risk-based logic: the higher the stakes of what a system decides, the more scrutiny required before it goes live. Hiring tools, credit scoring, medical diagnostics, law enforcement applications: all classified as high-risk, all requiring conformity assessments, bias audits, and documented human oversight before deployment. The underlying argument is not complicated. If a system makes decisions that affect people's lives, someone needs to be accountable for how it was built and what it gets wrong.
None of this needs to be slow. It is slower than shipping the first output that looks good in a demo. It is considerably faster than rebuilding trust after the system does visible harm at scale.
That is what responsible AI means in practice. Not a value statement on a website. The design discipline that makes the system work for everyone it touches, not just the people who built it and the metrics they chose to measure.
–
P.S. Here a tip for active inclusion: The Tivi list recognises 50 women. Go through it, follow the people on it, invite them to speak at your events and explicitly ask different people to participate. The systems we build reflect the choices we make about who is in the room when we make them.
