Making a virus more contagious, or stretching the gap between the moment someone becomes infectious and the moment they show symptoms, does not require inventing anything new. Kenneth Mulvany says both are already possible with the technology that exists.
The public argument about AI danger is about models becoming autonomous and turning on people. Mulvany, who built a drug-discovery company on the technology and now sits on UK government panels about it, says the thing worth worrying about is people misusing it.
"I don't feel the existential threat that we're seeing so much in the media. Well, maybe I do, but there's more homegrown things that I'm worrying about, which is the misuse of this technology rather than this technology becoming autonomous and having a grudge against humans."
Mulvany founded BenevolentAI, and before that built and sold a Cambridge drug company to a US pharmaceutical buyer, so his view of where British capital runs out is a transaction he did himself rather than a policy opinion.
The full interview is covered here so you can skip it.
Here are the 8 insights that matter.
👤 Guest: Kenneth Mulvany, Founder of BenevolentAI and the keynote speaker at Cambridge Tech Week
🎙️ Hosts: Caroline Hepker and Charles Capel, presenters of Bloomberg Daybreak Europe
📰 Published: 14 September 2026 on the Bloomberg Talks podcast feed (Bloomberg Radio)
🔴 YouTube | 🟣 Apple Podcasts | ⏱️ 12 min
Key Takeaways
The bio risk is not a new pathogen, it is an edited one
A longer gap between becoming infectious and showing symptoms is the attribute he names
The threat he takes seriously is human misuse, not a model with a grudge
Frontier labs started shutting this class of request down about a year ago
Britain funds seed companies and global-scale rounds, and starves the stage in between
He sold his own Cambridge company small because the growth capital was not there
The US buyer took the products to market and made a lot of money on them
An exit fee on companies that leave would make Britain less competitive, not more
What Britain needs first is enough exits to recycle money back into the ecosystem
He has comfort in the frontier labs, and no visibility into how they work
1. Misuse, Not Machine Malice
The conversation opened on that morning's news that Anthropic and OpenAI had promised measures to rein in their most advanced models, including third-party oversight inside the businesses. The host asked what an AI entrepreneur made of it.
Mulvany's starting point is that the technology is old and the capability is not. AI, he said, has been around for a very, very long time, but the power of the current models poses a set of threats that did not exist five or ten years ago.
The specific area he named is biology, and the reason is that two things are moving at once: the models are becoming more powerful and more accessible.
He is explicit that the risk he loses sleep over is the one involving people: "I don't feel the existential threat that we're seeing so much in the media. Well, maybe I do, but there's more homegrown things that I'm worrying about, which is the misuse of this technology rather than this technology becoming autonomous and having a grudge against humans."
The same capability, he said, "creates enormous opportunities for us all" while changing the risk landscape, and he has been watching that second half closely.
2. You Don't Need a New Virus
The host put the scenario people reach for when they talk about AI harming or killing humans — a model creating a virus or otherwise interfering with biology — and asked, given that BenevolentAI works in biopharma, whether his own researchers could use more regulation or more room to understand what the AI is doing.
Mulvany's answer moved the bar down rather than up. "It doesn't take a lot. It doesn't necessarily take creating a new virus."
The mechanism he described is editing an existing one: "You can change attributes of a virus, make it more contagious or have a longer latency period between when you are sick and infectious to when you start showing symptoms. And those are all within the realm of today's technology." A longer latency period means more time spent spreading a disease before anyone knows they have it.
The labs have already reacted. "And it's probably now maybe a year ago they began to shut some of these things down and look at this much more closely."
He cited a disclosure from one of them as evidence. "I know that Anthropic published something not too long ago where it elicited the number of things that it has shut down where it felt that it was beginning to get dangerous."
3. What Benevolent Did in COVID
The counterweight he offered is the same capability pointed the other way, and it is work his own company did.
During COVID, he said, BenevolentAI identified a treatment for patients who were seriously ill and hospitalized. "It took about 48 hours for it to do that."
The bottleneck was everything after the discovery. "I mean, it took a very long time for it to get through its clinical trials" — about a year, on his account, to be tested and ultimately approved.
That is the asymmetry he is pointing at: the search took two days and the approval took a year, which is the speed the company is now trying to bring to other pandemics.
He said BenevolentAI is working increasingly with governments on this, and that government is looking increasingly hard at it — with his company, the AI labs the host had named and others part of the same consortium.
4. The Missing Growth Capital
Asked about the spin-out landscape at UK universities, and about the government's stated ambition to double the number of British unicorns, Mulvany answered with a map of where money is and is not available. He has, he noted, roughly 20 years of personal experience of it.
"If you're an early startup, I think that there's money there."
At the other end, for a company raising to expand at scale, he said the market is relatively satisfied, and that this is probably more true globally than in Britain specifically.
The hole is in the middle: "But if you're a growth company, there's precious little capital."
The host's framing of the same problem was where the money comes from. Scale-up money, she said, arrives from the US or the Middle East, and "that leads to a slow drift of the UK's most groundbreaking startups and fastest growing companies out of the UK itself."
5. He Sold Small, and Said Why
The section of the interview that gives the rest of it its weight is Mulvany describing his own exit as a case of the problem he is diagnosing.
He used to make the opposite argument in public. He was a big advocate of saying that "you'll never sell yourself as a big company if you sell yourself as a small company," and he kept saying it publicly until the day he sold his own Cambridge company as a relatively small one.
He put the price at about $500 million, and gave one reason for it: "the market really wasn't there. I wasn't able to expand." Then: "And so I was bought by a U.S. pharmaceutical company."
What happened next is the part he returns to. The buyer took those products to market and made a lot of money on them.
His previous company was ProximaGen, and its shareholders did well — but he frames that as the smaller outcome: "But if they had the risk appetite, they could have done exceedingly well. That would have been returned into the ecosystem."
6. Procurement as a Leg Up
Mulvany moved from the diagnosis to a remedy without being asked for one, and the remedy is government purchasing rather than government money.
If he were in government today, he said, the first thing he would work on is the procurement cycle — and not only for medicines, but across the board.
The argument is that some companies need a leg up in order to begin competing at all, and that buying from them is the way to give it.
The obstacle is the process itself: "And it's very, very difficult to work your way through UK procurement, UK government procurement."
7. No to an Exit Fee
The host asked directly whether there is a case for charging companies a fee when they move their business overseas. Mulvany said no, and gave three separate reasons.
He said flatly that he is not a fan of making it harder for UK companies, then took the idea seriously enough to ask who would collect: "I can see the allure in that. I don't know who would get the exit fee." Would it go to the shareholders, or to the government?
The second reason is the effect on the companies still there: "I think that may make it less competitive."
The third is that a company with growth potential whose own domestic shareholders will not fund it has a right to go elsewhere. "I think that it should be allowed to do that unfettered."
"I may be simplifying it in many ways, but I wouldn't say that an exit fee would help. I think it would just make it less competitive."
8. Comfort, but Not Certainty
Asked in closing whether he is convinced that the big AI firms have life-sciences and biological risk under control, Mulvany drew a line between what he can see and what he cannot.
"I don't know all the interworkings of what happens. I do sit on government panels. I do think that there's transparency amongst those panels and I have comfort that the frontier models are doing everything that they can do at this particular point in time."
He then raised the obvious alternative and dismissed it as impractical rather than wrong. "You know, kind of the notion that, okay, we'll pause this. That's tricky, isn't it?"
The host agreed that it is, and the interview ended there.
Bonus Insights
The host's setup for the whole conversation was a set of British contradictions, delivered at the top: the Chancellor wants to double the number of unicorns, "The UK's got three of the world's top ten universities and lots of spin-outs," and yet experts describe "a persistent technical literacy gap" and mature spin-outs are "increasingly leaving the UK altogether." The interview was recorded on the opening day of Cambridge Tech Week, where Mulvany was the keynote speaker.
On the unicorn target itself, the host noted that the government has attached no deadline to it. "They haven't actually put a date on that. They just said they want to double it. So we don't have a timeline yet."
The host brought up an essay Dario Amodei published over the weekend as the reason the questions have multiplied — specifically the idea of coordination between democratic and authoritarian governments, and of outside people holding employee-level access to inspect the models. Mulvany was not asked to respond to it directly; the host used it to pivot to the UK.
On why recycled capital matters more than rules, Mulvany pointed at the United States as the working version of the system. "What I think is that first, the country needs quite a few successes under its belt for that money to kind of return to the ecosystem." Then: "That money has returned on the investment side early. They become big companies, people exit, they invest that money, and so the circle continues."
Mulvany's bottom line is that the two subjects he was asked about have the same shape: the danger in AI is what people choose to do with a capability that already exists, and the weakness in British technology is not a shortage of ideas or of seed money but a missing rung between the two, which he would fix by buying from those companies rather than by fining the ones that leave.
Products, Companies & Tools Mentioned
BenevolentAI (Mulvany's company, which works in biopharma, found a COVID treatment for hospitalized patients in about 48 hours, and is now working with governments on pandemic response and misuse risk)
Anthropic and OpenAI (The two labs that had just promised measures to rein in their most advanced models, including third-party oversight; Mulvany cites Anthropic's published count of requests it shut down as dangerous)
ProximaGen (The Cambridge drug company Mulvany built and sold to a US pharmaceutical buyer, in his telling for less than it was worth because the growth capital was not available in Britain)
Cambridge Tech Week (The event that opened the day of the interview, at which Mulvany was giving the keynote)
If this was worth your time, send it to someone closer to the industry than you are.
Get the latest market chatter as it happens: