Nov 6
/
James Kavanagh
Taking the Leadership Leap in AI Governance
You don’t need a title to lead. You need a signal worth acting on, and courage to mobilize others around it.
I wrote an article earlier this year about the three career leaps in AI Governance, where I told the story of my own story of practice and leadership leap at Microsoft and AWS. Like many professionals, I started by focusing on expertise as an individual with no direct team. But within a few years went from a small team in Canberra, to London leading more than 45 senior people across 25 countries, and influencing or making significant strategic decisions, including building Amazon Web Services first Responsible AI Assurance team. On that road, I had to learn to lead through influence, ideas and culture rather than my own expertise.
But there's a subtle aspect of that career story, relevant for anyone aspiring to a leadership role in AI governance. What if you don't have a team? Can you take the leadership leap without the title?
WIth absolute certainty, yes, you can. And in AI governance, it's one of the essential skills.
It all comes down to what you mean by leadership. Ron Heifetz's writing on adaptive leadership draws a really clear line between leadership and authority. As he frames it, authority is the formal power you're given: a title, budget, people who answer to you. But he describes leadership as an activity. As the work of mobilizing people to face hard problems and make progress on solving them.
You can hold plenty of authority, and exercise no leadership at all.
And you can lead with nobody reporting to you.
What our data shows
We now have more than 600 practitioner results from our free AI Governance Assessment. It's a challenging assessment, ranging from novice through to world-leading mastery in a set of five dimensions. Now half of those taking the assessment score at Novice overall, which is what I'd expect in a profession this new.
When we look at each of the individual dimensions – Knowledge, Application, Resources, Influence and Judgment – we see that 39% are at a Developing level or higher on Knowledge, while only 22% are at an equivalent level on Application. Again, not really surprising that there's a wide gap as it's harder and slower to put theory into practice.
Two of the other dimensions describe leadership in exactly Heifetz's sense. Influence is getting engineering, product, legal and leadership to act when you have no direct control over them. Judgment is reading the context and calibrating your response, including telling a technical challenge from an adaptive one. On both, less than 25% of practitioners assessed their capability at a Developing level or higher.
Now that did surprise me. Most of the people I work with in AI governance are mid-career, many moving across from law, security, privacy, audit or engineering. They've spent a lot of their professional life getting things done through people who don't report to them. So why is it that so many rate themselves as beginners on influence and judgment?
We dug into this by talking with people and watching how they progress through our programs. And in most cases, I perceive the skill is usually there. What seems to be missing is the recognition, or perhaps the confidence, that it counts in a different field. I think a lot of people imagine an AI governance career as a straight line: learn the domain, get your certifications, get a job, then practice it, and only after a few years earn the right to lead. So they set aside or diminish their confidence in the influence and judgment they already have, as if it belonged immovably to a previous career.
It doesn't. I think it may be the most transferable thing you bring. What it needs is practice in this new context: making decisions, convincing people, debating the hard calls and that does need to be supported with the right knowledge.
So what does leading without authority actually involve? I think it comes down to three things. You notice a signal that nobody else owns. You mobilize people to act on it, even though they don't report to you. And you accept that doing so carries real risk, but you do it anyway.
Start with the signal
The first half of leading without authority is noticing. Karl Weick's word for this is sensemaking: how people pick up weak, ambiguous cues, make sense of them, and turn them into a case others can act on. Most organizations are full of signals that nobody owns. And I think leadership often starts when one person decides to own one.
Joy Buolamwini is a good example in our field. When she was a graduate student at the MIT Media Lab, she built an art installation with face-tracking software. But it wouldn't work, it simply counldn't detect her face - until she put on a white mask. She wasn't the first person to notice that, plenty of people did and worked around it. They moved on. She treated it as a signal.
"How can I know what I think until I see what I say?”
- Karl Weick, Sensemaking in Organizations
- Karl Weick, Sensemaking in Organizations
So she founded the Algorithmic Justice League in 2016. And two years later, her Gender Shades study with Timnit Gebru showed commercial systems from IBM, Microsoft and Face++ misclassifying darker-skinned women at error rates up to 34.7%, against no more than 0.8% for lighter-skinned men. IBM and Microsoft both moved to improve their systems. She had no authority over either company.
My own signals were less dramatic. At Microsoft in Australia, government customers were telling us they couldn't move sensitive workloads to the cloud without far more resilience and security than we could offer. So I led the case and built the Australian Azure regions in Canberra, specifically to meet that need. At AWS, my job was regulatory intelligence, building the platforms we used to understand the world's regulations. The signal that kept rising through that work was responsible AI. So I led the case and built the AWS Responsible AI Assurance team. Nobody handed me either problem. Each one started with taking a signal seriously.
“If you have a face, you have a place in the conversation about AI.”
- Joy Buolamwini, 2023 NPR Fresh Air interview
- Joy Buolamwini, 2023 NPR Fresh Air interview
Then mobilize
For me, the second half is mobilizing. In The Practice of Adaptive Leadership, Heifetz defines adaptive leadership as "the practice of mobilizing people to tackle tough challenges and thrive". Once you've made sense of a signal, the work is getting people to act on it, even though most of them don't report to you.
At Microsoft, that meant building the business case, bringing government customers on board with the idea, and winning support from other leaders inside the company. Eventually we committed the capital and built the capability in partnership with Canberra Data Centres. At AWS, it meant leading the push to establish regulatory assurance and pursue ISO/IEC 42001 certification, and then achieving it. I had a team at AWS, and authority over their work. But the leadership wasn't in the org chart. It was in seeing the need and moving people well beyond my own team in engineering, security, legal, data science and others to meet it.
Katalina Hernandez shows the same pattern. Her background was law, privacy and compliance, not AI. Rather than wait for a role to be created, she stood up the Responsible AI function inside Vodafone Intelligent Solutions, took a seat on the Group AI Board, and ran EU AI Act implementation from the inside. Nobody gave her that mandate, she created it. You can check out my podcast conversation with Katalina that goes deeper into how.
And Christina Wiremu-Brook, who'll be joining us on a future episode of the podcast, shows what mobilizing looks like across disciplines. Her background spans leading an Indigenous education non-profit, consulting and data science. As the AI and data specialist leading Safe AI at the NSW Department of Education, she built AI ethics and governance frameworks to enable safe AI in education. She calls herself a socio-technical translator: someone who can engage executives, shape social policy and run the analytics herself. She is now founder and CEO of Kuria AI, and this year she won the Women in AI APAC award for AI in Cyber Security (Risk Management).
I think this act of mobilizing others is also where leadership starts to last. The test I use is whether what you built keeps working after you step away: the mechanisms, the habits, a culture where people raise concerns. That's what turns one person's push into an organization's practice.
Leading this way has a cost
“To lead is to live dangerously.”
- Ronald Heifetz, Leadership on the Line
- Ronald Heifetz, Leadership on the Line
Now I'd be doing you a disservice if I made this sound safe. Heifetz is blunt about it in Leadership on the Line. Leading means asking people to face losses they'd rather avoid, and they tend to push back on whoever is asking. Without formal authority, you have less cover when that happens.
Sometimes the work exposes a difference in values. Timnit Gebru co-led Google's Ethical AI team and co-wrote a paper on the risks of large language models. Her employment ended in December 2020 after managers asked her to withdraw it. She says she was fired; Google said she resigned. In May 2024, Jan Leike resigned from leading alignment at OpenAI, saying safety culture and processes had "taken a backseat to shiny products".
I've been through a version of this myself. I left Amazon partly because of a difference in values about where AI is heading, and my own belief that Big Tech needs to invest more and take more accountability for safe, secure, lawful operation of AI.
What strikes me in so many leadership stories is that leaving didn't end the leadership. A year after Google, Gebru founded the Distributed AI Research Institute. I went on to found AI Career Pro, and proud to help practitioners build their skills and lead. The action you mobilize inside an organization can be the seed for far greater momentum outside it.
Find the courage to lead
So my advice - take honest stock of how much you bring. If you've spent years persuading people who don't report to you, making judgment calls with incomplete information, or moving a hard decision through an organization, then I think you already have the raw material of leadership. It doesn't stay behind when you change fields.
You might have to find the courage to use it before anyone gives you the title. And frankly, in AI governance, you'll probably never have a large team. But you'll still need to move engineering, product, legal and executives, often all at once. Learning to lead without authority is part of the work from your very first day.
And if you want to see where you stand on Influence and Judgment, the AI Governance Assessment is free, and you'll find it here: assess.aicareer.pro.
You don't need permission to lead.
You need a signal worth acting on, and the courage to act on it.
Subscribe to our newsletter!
Our Doing AI Governance newsletter features the latest in AI Governance news, research and expert insights.
