# Bill Gates: AI Risks, Cyber Threats, and Economic Shifts

**Podcast:** FT Tech Tonic
**Published:** 2026-11-04

## Transcript

Hello, and welcome to Tectonic from the Financial Times.
I'm Madhumita Merjia, the FT's AI editor.
Over the next few weeks, we'll be releasing episodes exploring the future of artificial intelligence, finding out where the technology is going, how the industry is changing, and what all this could mean for businesses, the economy, and the wider world.
We'll be speaking to people working at the forefront of AI.
experts in the industry, and some of the most influential voices in technology.
Starting this week with Bill Gates.
Bill Gates is, of course, the co-founder and former CEO of Microsoft.
Today, he's chairman of the Gates Foundation, his charity that spends billions every year on healthcare and fighting poverty around the world.
He's been enthusiastic about AI's potential to help tackle some of the world's biggest problems.
But recently, he's become one of the leading voices.
sounding the alarm about the potential dangers of AI, like bioweapons and hacking, and calling on governments to act urgently.
I spoke to Bill as part of the FT's Future of AI Summit, which took place this week.
I started by asking him about the rapid development of AI over the last few years and why that has now sparked so much concern.
Well, starting with ChatGPT, which I got to see about nine months before it was released.
The improvement's been quite amazing, but the most stunning thing is that starting late last year, the coding capabilities, starting with cloud code, became better than any human being.
And that's great for writing software.
It's also a bit dangerous in terms of being able to find cybersecurity flaws.
So we crossed a very...
amazing threshold.
And since then, the models have just gotten better.
In fact, you know, they have new versions that they haven't released that'll probably come out before the end of the year and then two major releases next year.
So it's stunning how capable these models are and the potential for good, the potential for harm can't be overstated.
You've been really unique in that respect, though, in that.
amongst your peers of technologists and leaders in this field, you have been a voice of warning and, you know, waiting people up to the sort of outcomes and the huge risks from these models.
Tell me what...
prompted you to write so persuasively over the last few months about the risks of AI.
We can see why the capabilities can bring a lot of good and the sort of applications and the industry.
But what is it about the risks that have really made you use your voice in this way?
Well, I'd say the people at the AI companies, whether it's Sam or Dario or Demis, Mustafa, they've also been very clear that they're concerned, but somehow their voices get discounted that, you know, they're exaggerating because they want to promote their product.
In my case, although I'm from the industry and I get a chance to spend a lot of time with these guys, I'm a little separate.
And so I think my speaking out with a loud voice hopefully gets the rest of society to start engaging.
You know, it's so easy to say, This is a technology like those from the past where the concerns were overblown and the dynamic marketplace adjusted.
Here, though, the technology is utterly different than anything we've seen before.
And so, you know, I mostly like to speak out about more money for malaria, saving children's lives, the work of the foundation.
I decided I really needed to speak as strongly as possible.
about the need to minimize some of these dangers and a lot of work that's required to take advantage of these advances in areas like medicine or education or how the government itself is run.
So you talked about, you know, the fact that the voices from within the industry have been in some ways ignored when they talk about risk.
And I would say partly maybe the reason is that they are also full speed ahead developing the technologies and implementing them worldwide.
So there's a little bit of a tension there, you know, if they truly believed in the risks as you've described them as being, you know, potentially able to kill billions of people, you would expect that they would do something to slow themselves down.
So I'm curious your perspective on what is the most misunderstood thing about the technology today by those outside of the AI and tech industry?
Well, it's so stunning to say there'll be an intelligence better than human beings.
I just don't think people can get their mind around it.
I use analogies like the aliens came, but we invented them on our computers or that, you know, evolution, as you went from, say, monkeys to human beings.
It's of that order of magnitude.
It should not be compared to, say, the internet or the PC, the technologies that I was lucky enough to be involved in.
And so I think it's going to take a lot.
Also, people saw the early versions were so limited and it was easy to say, okay, there's some human element that will never be achieved by these models.
But in terms of science or even business activities, that's not the case.
You know, the next year we'll see.
such incredible improvements that the rest of society has to both believe this and start debating how we're going to adjust.
So I want to spend a moment on two examples that you've talked about before, and that seem clearly to be sort of the major near-term risks.
cybersecurity and bio, right?
Both areas that you have worked in, you have pioneered and spent a lot of time thinking about.
Can you talk about the kind of dark side of that?
What is it about these capabilities, particularly in cyber and bio, that can cause major near-term risks to our society?
Well, in cyber, you know, we live in a very digitized society.
You know, hospitals, the electric network.
banks, they all depend on a very complex set of software, which the offensive capability of AI models to find flaws and therefore either shut down or scramble those digital systems, that capability exists.
If you take what's called the red team and attack team and assign them to go after those things, if they have the latest AI tools, particularly the open source where you take out all of the monitoring and safeguards, then you will be successful at those attacks.
Now, over time, you can use the AI to redo that software, but that's on the order of five years of work to do all those replacements, even if we...
are very diligent and make it a top priority.
And because there's a business model for using crypto to, say, pay me a ransom, you're going to see a dramatic increase in cyber attacks in the next few years.
You know, the open-weights models, you can take out all of the monitoring, so you won't even see it coming.
And what about bio risks?
Are you particularly concerned?
Okay, bio is, you could say, even scarier because the scale of the damage, you know, you could literally have something that is significantly worse than COVID.
Now, bio, there's not a business model for.
So actually predicting when somebody would do that is more difficult.
And I'm hopeful we get lucky on that one.
But there's a lot of things.
having to do with the safeguards and the models, a lot of surveillance so we could catch something very quickly.
There's a lot we need to do that we're just not doing.
So the work, the writing that you've done over the past few months has been described as a sort of call to arms.
And I, as well, having read it, you know, I really appreciated how strong you are in what you call for.
There's no sort of...
equivocating in what you're saying.
And we see a lot of that in the industry today.
It's sort of like if this or that.
But I would love for you to lay out what do you think needs to happen now?
You've laid out what the risk is and we're all starting to realize, you know, the power of the models.
What needs to happen now over the next, you know, six months to make this go well for us?
Well, each company.
When they were created, they had a notion that this technology was different.
You know, OpenAI had a clause that said if it passed AGI, they could ignore the capitalistic tendency to put it out there and just not ever release it.
The DeepMind founders negotiated with Google to have some special treatment.
But as soon as we had multiple companies...
Then the only way to solve this moves up to the government.
You know, it has to decide is open weights okay.
And when you have a new model, do you have a benchmark about the monitoring and the safeguards that are in that model?
And so far, the U.S.
doesn't even have mandatory review.
And so other governments, including China, there's not much value.
in acting unless the U.S.
is part of that.
So we're kind of stymieing.
You know, I think the U.S.
will come along on this.
The only question is how many bad events need to take place before they change their posture.
Yeah, well, exactly.
You've captured the problem here, right?
We're in this total stasis.
And the reason is that the only government that can lead on this is the U.S.
government, because this is where Of course, we have the open-week companies in China, but really the leading technology is in the U.S.
and on the West Coast of the U.S.
But, you know, the current administration has shown very little proclivity towards any sort of regulation and, in fact, has come out clearly and said that there will be self-regulation.
You know, we've seen recent meetings at the White House.
advice would you give them today, say two or three tangible things that they could be doing today to shepherd the technology safely and to control what's coming out of the labs in the U.S.?
If dangerous models with safeguards removed are continued to be allowed, then the next defense is to redo your software to get rid of the security flaws.
That's very hard, but the sooner we start, the better.
On the biosecurity front, it's to have massive detection where you're looking at air, sewage, blood, and trying to catch infections when it's still affecting 1,000 or 10,000 and not let it get to 1 million, 10 million, 100 million.
You know, that kind of surveillance can be done and the cost would actually be reasonable.
And so there are people who are looking at how do we pull this together?
That would need to be a global effort and it would have a lot of benefits.
So in terms of the technology side of how we apply safety and, you know, you've talked about a supervisory layer.
Dario Amadai has described it as pacing the frontier.
Nobody seems to really want to pause or stop.
But they have talked about this pacing, which I see as a kind of, I don't know, like a thoughtful, cautious proceeding of the development of the technology.
What do you think, both from the technological perspective and regulation, what would pacing look like in a way that actually works?
Well, the closed models have safeguards built into them.
We could do a lot of third-party auditing to see if they're as strong as they need to be.
There are oftentimes where you can get around them what's called jailbreaking.
And people like the UK Safety Institute are an example where capacity is starting to be developed to look at the safeguard capabilities.
This issue of the models themselves being out of control, that today is not the imminent threat.
But if you look at...
the versions that are coming, then that becomes a serious thing.
And there's just a lot of technology work to be done to change either how we teach these models or the basic structure underneath or how we encapsulate them.
The control problem, although it's not imminent, we need to put a lot more work into that and make sure that develops before next year's models come out.
And again, without government involved, it's very hard to see how that takes place.
Yeah.
On that question, the self-regulation question, you have made it clear before that you don't think that would work in the context of, you know, huge risks.
But given that's what we have today, I'm curious about your take on Do you believe that we can have some sort of self-regulation, at least until you have government regulation coming in?
And with the three main competitors we have out on the market, you know, is that going to work?
And do you think that can actually slow it down enough that we can build these layers and safeguards around technology?
No, I don't think that's the right way to go.
There's just too many participants and their competitors.
So I think it's great that the industry is being so open.
You know, people say you need to get better PR as an industry, you know, when people are being frank about these things.
So I give them a lot of credit for the way that they're speaking out.
You know, it's a very novel situation that it's the closer you are to the development, the more you're saying, hey, we need safety regulation here.
So eventually we'll have incidents that will lead.
to a government response.
I was going to ask that actually.
What do you think would happen before it would force even a unilateral slowdown of a single lab?
We had a sort of early incident like that with Hugging Face being hacked by OpenAI.
And I think that's caused huge changes internally in how these labs are thinking about safety, even at the training stage of the models.
But what would be the threshold to cross before it would force a unilateral slowdown of what these labs are building?
I just wouldn't expect a unilateral slowdown.
The urgency of work on control was highlighted by hugging face.
And everybody's almost overly focused on that now.
It's super important.
But it's really the bad people.
Using the models where you remove the safeguards, that is the near-term challenge that we need to face.
Over the next several years, that control issue will be added in.
And that's at the same time that you have things like, how should kids be able to use it?
How do you deal with the job market effects that this has?
And on the good side, how do you accelerate?
get rid of NHS waiting lists?
Can it discover new drugs?
None of the good stuff happens naturally.
And through my work at the Gates Foundation, you know, really driving this to help in health, to help farmers who get very little advice, you know, that's the positive side.
And there's so much positive work to be done.
I want to talk a bit more about that.
I think particularly AI in medicine is also a passion and interest of mine.
And I would love to hear your take on it from the front lines of having implemented this in parts of the world that need it.
But before that, you touched on this briefly in your comment then, but the question of jobs, that's the other kind of major near-term risk that you've written about very deeply and thoughtfully.
And I would, you know, there's been so much debate, data, you know, economists.
fighting over this, technologists jumping in about the question of, are we going to see loss of jobs?
Will they just be replaced eventually?
You know, like what happened with previous kind of industrial revolutions.
Can you set out your thesis of what you think is coming with regard to change in job markets and who will be impacted most in the coming years?
Well, I say very strongly that past technological advances are no guide to this.
That will just mislead you if you look at those.
This is a situation where you're exceeding human cognition.
So your ability to be a telesalesperson or to be a follow-up person after medical care who talks to patients or to work in an accounting department, the AI, once it's properly set up, And as we get these much, much better versions, it is very cheap to replace the white collar labor.
And there isn't some other area.
You know, it's not just health.
It's not just law.
It's not just business consulting.
It's every area of the economy.
And so the taxation policies that we have.
And the fact that we make labor is where you put the tax burden.
And we're really generous on the capital spending to say buy the robots or buy the tokens.
That is going to accelerate these replacement efforts.
And it's, you know, you'll want to fund some safety net resources.
So only programming is at the point where the AI is clearly superior.
But many white-collar jobs will be added over the next several years.
I'll move to this question of what you're excited about, particularly in the area of health.
You know, where do you see AI really bringing positive change from the work that you've been doing?
Well, the Gates Foundation already has an ultrasound that using AI in the poorest countries of the world, we can do a scan of a...
pregnant women and determine if the delivery is going to be difficult or not.
So the woman knows, okay, do I need to get to a health facility where I can have a C-section if that's necessary?
So the ability to diagnose, it's just phenomenal.
It's very, very inexpensive.
Likewise, in our drug discovery work, we need new drugs for tuberculosis or malaria.
We have very ambitious goals to actually eradicate malaria.
So the AI tools are accelerating that work.
You know, given that all our money will be spent in the next 19 years, the fact that we can go about three times faster in our discovery allows us to be even more ambitious.
And so we're taking AI both on the science and also on talking to the patients so we can take somebody.
living with HIV or pregnant women and in their language, which right now the African languages are, the quality is very poor, but we announced a plan with 60 partners over the next several years to really get high quality for all of these languages.
So there, it's like having a virtual doctor or if you're a farmer getting top-notch advice about weather and seeds.
And so that our pilot programs in agriculture and health are just fantastic.
Thank you.
I want to tap into the aspect of your expertise we haven't talked about yet, which is you as a corporate technological leader.
We've just seen Anthropics S1 documents.
I'm not sure if you've yet had a chance to look through, but if you have, I'm curious what your take is on what this tells us about the market and the industry, given what you're seeing.
whether kind of the numbers or the market that they're targeting, kind of looking at the case they're making.
What does this tell us people on the outside of the industry about what to expect from these companies in the coming months or years?
Yeah, I think Anthropik's done a very good job talking about the risks.
They've always been very open, and this is another example of that.
Open Air and Anthropik are both doing a fantastic job.
moving the technology forward.
And the demand for this intelligence will be very high.
But there's also a lot of competition.
So the exact forecasting of, you know, what will the revenue be a year from now, two years from now, three years from now, that's very difficult.
I'm hoping these companies do well because I think they're thoughtful and the owners.
plan to be very philanthropic.
You know, for OpenAI, they've got their foundation.
Anthropics set shares aside and their founders have all committed over 80% going to philanthropy.
So I'm also talking with them about, OK, let's work together to make sure those dollars have huge impact in many areas, including global health.
You set up my final question perfectly, Bill.
Thank you, because that was exactly what I wanted to ask.
You know.
By all accounts, wealth inequality is just getting worse.
And AI has the potential to maybe change that either for better or worse and really could widen that gap depending on how people access it.
But also we are going to have this huge influx of wealth coming out of people who are making money from these AI companies, particularly now with the IPOs.
As you said, both the companies have foundations that they will fund.
What are your thoughts on what your fellow billionaires should be doing with this money?
Where will it work hardest?
And how can they work to remedy any inequalities born out of this new revolution?
Well, of course, the primary responsibility for dealing with inequality will always rest with governments.
We can't expect philanthropy, which is voluntary in nature, to be the safety net.
So in a world of AI, There will have to be changes in tax systems, including the ability to tax people with huge wealth, the ability to tax billionaires.
But whatever's left over after those those new taxes are put in place, I do think we're seeing a very enlightened group there at a very young age already asking the questions about, OK, what's the most impact I can have?
How do we be complementary to whatever government aid there is?
It is a tough time where a lot of government generosity going to the poorest countries has been cut back.
You know, it's almost an emergency situation where the health care cuts have met.
This year, more children will die.
The last year, after 25 years of really good reductions.
And so there's great discussions going on with.
all of these founders about how they can have high impact.
Great.
Well, that's a note for us to end on.
And we hope to see lots of good work coming out of the many billions that will be made and that you will be kind of stewarding that or shepherding it.
Thank you so much, Bill, for joining us this afternoon.
I hope you've enjoyed that conversation.
And we certainly have.
It's great to talk to you.
Thank you.
That was Bill Gates.
speaking to me at the FT's Future of AI Summit.
Many thanks to Bill.
We'll have more episodes exploring the future of AI over the next few weeks.
Keep an eye on this feed for those.
For now, you've been listening to Tectonic from the Financial Times with me, Madhumita Mergia.
The show is produced by Edwin Lane.
Manuela Saragossa is the executive producer.
Sound design by Breen Turner.
Original music by Metaphor Music.
Flo Phillips is the FD's Head of Audio.