Bill Gates, co-founder of Microsoft and chairman of the Gates Foundation, sat down with Madhumita Murgia, the Financial Times' AI Editor, at the FT Future of AI Summit 2026. He says governments are not moving fast enough on AI risk, and that this technology is unlike anything before it: past technology cycles are "no guide" to what comes next.
This article is our own summary, with short quotes, from the video's automatic transcript. Watch the full interview above for everything that was said.
The short version
- Coding crossed a line. Gates says that starting late last year, AI coding, beginning with Claude Code, became better than any human. That is great for writing software and also dangerous, because the same skill finds cybersecurity flaws.
- Cyber is the near-term risk. Hospitals, the electric grid and banks all run on complex software. Replacing the flawed software would take about five years of work even at top priority, and with open-weight models that have the safeguards removed, he expects a dramatic increase in cyber attacks in the next few years.
- Bio is scarier. The scale of damage could be significantly worse than COVID. His answer is detection: catch an outbreak when it affects 1,000 or 10,000 people, not 1 million, 10 million or 100 million, by monitoring air, sewage and blood.
- Self-regulation will not work. There are too many competitors, so the only way to solve it is for governments to act. He notes the US does not even have a mandatory review of new models, and that other countries see little value in acting unless the US does.
- Jobs are different this time. Because AI is exceeding human cognition, it becomes very cheap to replace white-collar labor in every area of the economy. Programming is the only field where AI is clearly superior today, but he expects many white-collar jobs to be affected over the next several years.
- The good side needs work. The Gates Foundation already uses an AI ultrasound in the poorest countries to tell pregnant women if a delivery will be difficult, and its drug discovery work is going about three times faster with AI.
What he said, by topic
Why he is speaking up
Gates says leaders at the AI companies, including Sam Altman, Dario Amodei, Demis Hassabis and Mustafa Suleyman, have said they are concerned, but their voices get discounted because they are promoting a product. Being from the industry but a little separate, he hopes his voice gets the rest of society engaged. He says he would rather talk about malaria and the work of the foundation, but decided he needed to speak as strongly as possible about the dangers.
Why people misjudge it
He says it is hard to get your mind around an intelligence better than human beings. He compares it to aliens that we invented on our computers, or the jump from monkeys to humans, and says it should not be compared to the internet or the PC. He argues that people who saw the early, limited versions assumed some human element could never be matched, and that this no longer holds for science or business work. He says unreleased versions will probably come out before the end of the year, with two major releases next year.
Cyber and bio
On cyber, he says a red team armed with the latest tools, especially open-source models with monitoring removed, would succeed at attacks on critical systems. A business model for ransom payments in crypto adds to the pressure. On bio, he says there is no business model for an attack, so predicting one is harder, and he hopes "we get lucky." He wants stronger safeguards in the models and a lot of surveillance so something could be caught very quickly.
What governments should do
He says each AI company started with the idea that this technology was different. He gives the example of an OpenAI clause about ignoring commercial pressure if it passed AGI, and special terms the DeepMind founders negotiated with Google. With multiple companies competing, the decision moves up to government: for example whether open weights are allowed, and whether a new model is benchmarked on monitoring and safeguards. He thinks the US will come along, and the question is how many bad events happen first.
Safeguards and control
For closed models, he wants third-party auditing of the safeguards, since people can often get around them (jailbreaking), and points to the UK AI Safety Institute as an early example of capacity. He says models getting out of control is not the imminent threat today, but could become serious with coming versions, and that more technical work is needed before next year's models arrive. On the idea of a unilateral slowdown by one lab, he says he would not expect one.
Jobs and taxes
He says the taxation policies we have put the burden on labor and are generous on capital spending, "buy the robots or buy the tokens," which will accelerate replacement. He says you will want to fund some safety-net resources. He expects the primary responsibility for inequality to rest with governments, including changes to tax systems and the ability to tax billionaires, because philanthropy is voluntary and cannot be the safety net.
The upside in health and farming
He describes AI-assisted ultrasound that is very inexpensive, AI in drug discovery for tuberculosis and malaria with an ambition to eradicate malaria, and a plan with about 60 partners to bring high-quality AI in African languages. He says patients could get what is like a virtual doctor, and farmers top-notch advice on weather and seeds. He says that since the foundation plans to spend all its money over the next 19 years, going about three times faster in discovery lets it be more ambitious. He adds that none of the good stuff happens naturally.
AI companies, philanthropy and Anthropic's filing
Asked about Anthropic's S-1, he says Anthropic has done a good job talking about risks and that OpenAI and Anthropic are both moving the technology forward, but that forecasting revenue is very difficult with this much competition. He notes both have foundations, and that their founders have committed over 80% to philanthropy, and says he is talking with them about global health. He ends by saying government generosity toward the poorest countries has been cut back and that he expects more children to die this year than last, after 25 years of reductions.
What it means for your business
- Treat security as a project now. If attackers can use AI to find flaws, aging software and unpatched systems are the first things to fix.
- Do not plan from old technology cycles. Gates argues past advances are a poor guide, so test your own assumptions about which roles AI will do cheaply.
- Start with the work that is software-shaped. Programming is where AI is already clearly ahead, so expect pressure on that work first.
- Expect rules to arrive after an incident. He expects regulation to follow bad events, so build safeguards and audit trails before you are asked.
- Put AI where access is the problem. His examples, from ultrasound to language access to farm advice, are cases where AI delivers something that was too costly to provide before.
- Watch the tax and policy debate. He expects changes to how labor and capital are taxed, which affects hiring versus automation decisions.
Video details
- Title: Bill Gates Warns of AI Risks, Job Losses and US Regulation | #FTFutureofAI
- Published by: FT Live on YouTube, November 4, 2026
- Host: Madhumita Murgia, AI Editor, Financial Times
- Guest: Bill Gates, co-founder of Microsoft and chairman of the Gates Foundation
- Watch: youtube.com/watch?v=g3wyStng7_M
- Note: quotes and figures come from the video's automatic transcript and the speakers' own statements, and are lightly cleaned. Check the video for exact wording.
Want to put this to work?
We help owners turn AI trends into a plan, a product and a launch. Book a free call or see our services.