This week, I wanted to write a profile of Dr Jade Leung, who has just been appointed as the Prime Minister’s AI adviser.
I didn’t speak to Leung herself (though I did reach out) but have instead pulled this together through research and conversations with those who have known or worked with her.
It’s not a complete picture, but I enjoyed the challenge. I’ve always been partial to Radio 4’s Profile programme precisely because it doesn’t feature its subject, instead piecing them together from various sources. Let me know if you like this style and if I should do it again – the next obvious candidate would be whoever ends up being appointed as ARIA CEO.
When Jade Leung moved from Hong Kong to New Zealand with her family in 2008, the first thing she noticed was the air.
“I could actually breathe. For the first few days on the ground, I felt like I was chewing mints.”
But as time went on, she noticed other, less awe-inspiring things about her adopted home, like the divisions in society along class and ethnic groups. Different outcomes were expected for young people in the ‘wrong’ part of town, the ‘worse’ schools.
As she arrived at the University of Auckland, she was determined to use her time to do something about it. So she helped set up a not-for-profit called NCEA Campus, which recruited high-achieving high school graduates to tutor pupils in preparation for their exams.
She relayed all this to a crowd at TEDxYouth@Auckland in 2013. Only 19, she was already a reasonably confident public speaker, making her point quickly in a distinct Kiwi accent that seems to have faded in more recent appearances.
The organisation had done good work, she told the audience, but hadn’t reached the schools and pupils most in need. So she and her team went on a fact-finding mission, asking questions to the people involved in underachieving schools, getting to know the ecosystem. They then tweaked their approach to optimise the courses for accessibility.
This thorough approach has been a hallmark of Leung’s work ever since. She gets her teeth into subjects and works hard to figure out what the real problem is. Then she starts working on the solution.
“I was known as the girl who never slept,” she quipped during that early talk.
Last week, Leung was appointed as the Prime Minister’s AI adviser, a position she takes over from Matt Clifford.
The press release was sparse, noting simply that she will “work to position the UK as the leading nation to help unlock the benefits and prepare for the impacts of transformative AI”. She will report directly to both the PM and to Peter Kyle.
A few posts welcoming the news on social media added some extra clues as to why she had been selected.
Clifford himself welcomed his successor, noting that she brings “a tonne of experience in frontier labs, VC and government and will do an amazing job of ensuring the UK is an AI winner”. Steven Adler, who worked at OpenAI at the same time as Leung, called her a “legend of the AI security and governance field” and congratulated the UK. Markus Anderljung, director of policy and research at the Centre for the Governance of AI, excitedly shared three starry-eyed emojis – though he may be biased (he and Leung are married).
So how did this high-achieving student determined to drive social change become the top pick to guide the government’s AI approach in little more than a decade?1
After completing her engineering degree at Auckland, Leung did a Master’s in Environmental Policy at Cambridge, taking the top mark in her class. Then in 2016 it was off to Oxford as a Rhodes Scholar.
At this point, she was set on an environmental focus, with her doctorate attached to the Geography department. But at some point in that first year, something changed.
She approached Professor Duncan Snidal, a professor of International Relations and fellow of Nuffield College, to talk to him about a new idea for her thesis: AI governance.
“The first thing about Jade is she’s really smart,” Professor Snidal tells me on a call from his book-laden study in Chicago, recalling his first impressions of her. “It’s one of the nice things about Oxford: a lot of smart people around. And she’s also quirky, but she’s a very social person. She laughs a lot. She’s fun to engage with.”
Leung applied to shift to the Department for Politics and International Relations in order to pursue her new focus. In a letter recommending the change, written in the summer of 2017, Professor Snidal noted that the international governance of AI was “an emerging and important topic and I’m not aware of anyone else doing anything on it”. Though Leung did not have much background in IR, he was confident she would be a “quick study”.
A quick study indeed. By the summer of 2019, she had completed her thesis, supervised by Professor Snidal: ‘Who will govern artificial intelligence? Learning from the history of strategic politics in emerging technologies’. You can read the whole thing here, if you wish.
By observing how other general purpose technology sectors – aerospace, biotechnology and cryptography – had interacted with politics in the past, she proposed that one could make predictions about how the politics of AI would unfold.
States, she wrote, would increasingly seek to regain influence over the technology when faced with challenges such as competition with other countries or public concern about potential harms.
“Firms will find themselves increasingly constrained by the legislative environment, and more pressured to serve national defence and security interests,” she predicted.
At the same time as all this, she was also serving as the head of research and partnerships at the Centre for Governance of Artificial Intelligence (GovAI). This organisation was part of the ill-fated Future of Humanity Institute at the time, but later became an independent entity.
It was founded by Allan Dafoe, who now works on safety and governance at DeepMind. He is thanked three times in her thesis acknowledgments.
What’s not entirely clear to me from my research – the missing “rosebud”, if you will – is what made Leung switch to AI governance research. What was the lightbulb moment?
This is not exactly a guarded secret; I’m sure she’d be happy to explain it herself. However, it’s also not hard to hazard a guess.
Someone who was leading a charity at 19, who then goes on to study major environmental issues, is clearly drawn to big questions about how to make the world better. Perhaps even how to save it.
It feels inevitable, therefore, that she would be drawn towards questions of long-term risk – especially while studying at the same university where the Future of Humanity Institute researchers were helping to establish technical AI safety as a credible field. No doubt I’m oversimplifying and missing some key influences, but it is clear there was something in the Oxford water regarding AI and its role as a possible existential risk at that point.2
“I have the privilege every day of working with a small but growing community of people that are trying to tackle these challenges with as much thoughtfulness and urgency as we can,” she said during a talk at TedxWarick conference in 2020. “And I can tell you this much, that we certainly need a lot more help.”
Rosie Campbell, now the managing director of Eleos AI Research, first met Leung on a visit to Oxford in 2017. She was considering pivoting her own career towards AI safety, and had heard Oxford was the place to go, so was meeting people in related areas to get advice. She found Leung very impressive.
A few years later, Leung had left GovAI and begun working with Sam Altman. At first, she was a partner at VC firm Apollo Projects, an idea by Altman and his brothers to create moonshot companies through rapid seed rounds.
She then transitioned to a role within OpenAI as a governance and policy adviser and was then promoted to governance lead.
Campbell had also ended up at OpenAI, and joined Leung’s team in 2023 to work on “dangerous capability evaluations”, which is what it says on the tin: looking at what kinds of abilities it might be dangerous for models to have.
“Jade was incredible,” Campbell tells me via email. “She had a clear and inspiring vision for the work which involved trying to take a rigorous, empirical approach to measuring risks and using that to make decisions around mitigations and deployment. She brought a lot of energy to the team; she would push us hard and had a high quality standard, but in a way that was supportive and motivating.”
The girl who never slept was once again balancing multiple projects, but maintaining a thorough understanding of issues from the top-level to the nitty-gritty. “I was impressed by her productivity,” says Campbell.
Meanwhile, a team was being assembled. Under Rishi Sunak, the Frontier AI Taskforce had been formed as a “start-up inside government” and it was on the lookout for top talent.
Leung joined in October 2023, just a month before the taskforce would officially evolve into the AI Safety Institute. According to Time, she took a big pay cut to do so.
As chief technology officer at the organisation (which was renamed as the AI Security Institute earlier this year), she built up its capabilities to run cutting-edge research on model safety, bringing in new recruits and developing tests.
“We are currently, I would say, the best single place that you could go to to have your models evaluated across a range of frontier AI risk areas in a fairly robust and in depth way,” Leung told Time in 2024.
So we come to this summer, when Leung’s combination of experience seems to have made her something of a straightforward appointment. She’ll be continuing to work with AISI, splitting her time between there and Number 10. But what will her priorities be?
There’s the work of her predecessor to consider, of course. Clifford’s AI Opportunities Action Plan set out a framework that is focused on infrastructure, driving adoption and nurturing national champions.
But there could be work to do to ensure that plan doesn’t remain, as Alex Chalmers characterised it today, “a suggestion note to the rest of the government”.
Based on Leung’s previous work, Professor Snidal says another key consideration will be figuring out what the UK’s role is as a “middle power”.
“A lot depends on whether you think middle powers have any autonomy in the technological space. That is, can the UK create its own AI industry, in effect, that will come anywhere close to Silicon Valley or to China?”
That pressing question of sovereignty, and whether the UK could ever truly have it – or would even need it – looks set to be close to the top of the agenda. Her appointment announcement also makes clear that she’s there to help unlock the power of AI as it relates to the Government’s ‘Plan for Change’, so driving real-world adoption across key services will also be on the table.
Doing all of this safely, with what’s best for humanity in mind, must also be one of her chief concerns. Leung seems to be a person who acknowledges the scale of the challenge of AI, but is determined to work through it anyway.
“I think we were both motivated to work in AI because we could see that it would be a transformative technology with very high-stakes: the potential benefit could be huge but navigating the technical and governance challenges would be extremely difficult, yet essential to a positive outcome,” says Campbell.
“Jade cares a lot about making the future go well.”
Teatime scroll Each week I share links to writings, events, tweets and other conversation-starters. If you have something you think should be in here, feel free to reply to this email or DM me.
🚪 10 Downing Street’s Data Science team (10DS) is opening up one of its hackathons to the public for the first time ever, with a theme of rewiring the state. It’s at Imperial College London on Tuesday 16 September and top teams will be invited back to pitch their ideas at Number 10.
✴️ Across the pond, the launch of The Argument seems to have prompted plenty of, well, arguments. Have a read of Jerusalem Demsas’s opening essay about liberalism and decide for yourself.
🛋️ With apologies, last week’s edition had the wrong link for the ‘live near friends’ movement round-up. Here is the correct one.
🚲 Dara Ó Briain’s tweet brought attention this week to the way bridges between Hounslow and Richmond are becoming Checkpoint Charlie now that the two boroughs authorise different e-bike providers. It’s highlighted some downsides both of London’s borough-by-borough governance system and the competitive market for micromobility providers. Interestingly, this lack of co-ordination might be solved by the English Devolution Bill, which is due for its second reading in a couple of weeks. That’s because the London Assembly has called for TfL to be granted powers via the bill that would enable them to manage services on a city-wide basis. (Thanks Hounslow Cycling on X for pointing this out).
🔬For Research Professional News, Frances Jones reports that the Institute of Science and Technology has joined the voices criticising the troubled Alan Turing Institute. “[The UK is] lagging on education, legislation & safety in AI,” a spokesperson said. “This is critical work that should have been taking place within ATI.”
🏘️ Let’s all move to Maltkiln.
🛰️ Lots of disappointment over the folding of the UK Space Agency into DSIT. Loic Fremond has this useful analysis if you want to get a bit deeper.
💹 For any women working in VC, there is a meetup for aspiring female partners on 10 September in London. Starling founder Anne Boden is the guest speaker.
🖊️ We had a great time writing together at the very first London Generalists’ Writing Club! Join us for the next one on 1 September.
Yes, she’s only 31. I verified this by finding her on Companies House.
I left the year Leung arrived, so clearly I either just missed this or – more likely – my concerns were of a far more pedestrian, undergraduate, humanities student flavour at that time. The price of a summer ball ticket and Paper 4: English 1300-1550 come to mind.
Great read, thank you. I’m a fan of this style too and an avid reader of CEO profiles in the FT!