Earlier this month I attended World Summit AI in Amsterdam (7-8 October 2026). It was the 10th edition of the summit, and the theme this year was "Guardians of tomorrow". One theme was common throughout - who is responsible when AI acts on our behalf?
Who carries the responsibility when AI acts on our behalf?
Audrey Tang (Taiwan's Cyber Ambassador, and its first Digital Minister) suggested that we should be designing incentives and liability so that the companies building the technology have a reason to act responsibly. Her example came from another technology. When the world learned that CFCs were damaging the ozone layer, we did not give up our fridges. Public awareness and effective regulation changed the industry that made them.
Tang's suggestion is also about plurality built into the design - people deciding together, with technology that helps them do it. In Taiwan in 2024, after a wave of deepfake scam ads, the government sent text messages to 200,000 random numbers and chose 447 people to deliberate online, in groups of about 10, on how to solve this. Their proposals included ideas like making a platform pay for the losses when it shows an unsigned ad that turns out to be a scam. Parliament passed selected ideas within a few months, and deepfake scam ads went down by more than 94% over the following year.
What it means for Ownership
This brings me to Amazon's Ownership leadership principle:
Leaders are owners. They think long term and don't sacrifice long-term value for short-term results. They act on behalf of the entire company, beyond just their own team. They never say "that's not my job."
I wrote about this principle before, in Ownership: acting on behalf of the entire company with AI. The part I want to focus on in this context is "beyond just their own team". The people an AI system affects are often outside the team that builds it, and sometimes outside the company that deploys it - your employees, your customers, and in some cases the public.
What you can do as a leader rolling out AI
The first thing is to involve the people affected by the solution. If you want an AI rollout to work, ask the people who will use it how they see it even before the requirements are fixed, in small groups, and look for the ideas that different groups can agree on. It is a much smaller version of what Taiwan did, but it works in a similar way. Here are the steps I suggest:
- Map who is affected. Start with the team that builds or deploys the AI, and then move outward: the other teams whose work will change, your customers, and anyone outside the company who may feel the effects. For each group, write down what may change for them.
- Bring them in early. Invite people from each group into those small-group conversations, so what they say can still shape the solution, and it is actually co-designed as you are building together.
- Commit in advance to act on what they agree on. Tang did this as a minister, when she committed to bring anything that reached rough consensus to parliament. You can make a similar commitment about what goes into your solution.
- Give them part of the solution to own. People from the affected groups can help define what a good result looks like, and then test the AI against it before and after it rolls out.
- Keep the conversation going after launch. Give the people affected a simple way to flag where the AI isn't working for them, and let them see what changed as a result.
The same idea showed up in a session on business users building their own apps, where one of the lessons shared on moving from pilot to scale was to "build with users from the beginning". I wrote about that session in Build, ask, then finish.
The second thing is to look at what your incentives reward. The fridge example worked because it gave the companies involved a reason to change. Inside your organization, the same thing happens through what you measure and reward. As I wrote in Build one AI frontier team before you scale, it is tempting to measure how much people use AI, because it is easy to count. But a team measured only on that may be running fast in the wrong direction.
It is also critical to let people who use AI build their own capability along the way. Tang compares letting AI do our thinking for us to sending a robot to the gym to lift the weights: very impressive, "but our muscles don't grow this way." The same may happen inside a company. If AI does the thinking for a team, the team may still hit its numbers while its own capability stays where it was.
Tang has written about these ideas at length in Plurality, the book she wrote with E. Glen Weyl, which is free to read online.
Go deeper
If you want to dive deeper, watch Audrey Tang's keynote and conversation with Hala Hanna, the Executive Director of MIT Solve.
Your action step
Who in your organization (and beyond) is affected by the AI you are rolling out this year, and have they had a say in how it works? Follow the steps suggested above to define shared ownership with them.
If you want help mapping who your AI rollout affects and bringing them into the design, that is the kind of work I take on in AI strategy advisory engagements and in sessions as an AI keynote speaker and workshop facilitator.