Bill Gates is no longer framing advanced AI risk as a distant question. In a new interview with MIT Technology Review, the philanthropist and former Microsoft CEO says society has already moved past several points where stronger safeguards should have been in place.
His warning is not that AI has no upside. Gates remains clear that the technology can help in areas such as agriculture, health care, education, vaccine and drug innovation, and access to government or nongovernment support. His central concern is that the risks are arriving faster than public policy, government capacity and public debate.
The thresholds Gates says AI has crossed
Gates told MIT Technology Review that AI has passed danger thresholds in several areas: bio-capabilities, cyber-capabilities, psychosocial capabilities, job-market-destruction capabilities and signs connected to lack of control.
That list matters because each category points to a different kind of disruption. Bio-capabilities raise the question of what models can help design or manipulate. Cyber-capabilities raise the possibility that powerful tools could make attacks easier. Psychosocial capabilities point toward dependence on AI systems. Job-market concerns focus on whether low-cost AI can replace a substantial share of white-collar work.
Gates said he expected louder public concern before these thresholds were reached. Instead, he described being surprised by how little discussion is happening beyond the AI industry itself.
Any model that can make novel molecules should be monitored.
His strongest warning is about biological risk. Gates said any model able to make novel molecules should not be freely copyable into an environment where monitoring can be removed. He also said the US should define monitoring requirements and approach China to agree on them.
Why the jobs argument is different this time
Gates acknowledged the familiar argument that earlier technologies did not create a net reduction in jobs. He also said he has made that argument himself. But in this interview, he argued that AI is different because it can substitute for human cognition across many industries at modest cost.
He did not say every job is equally exposed. He pointed to roles that rely on complex implicit judgment, relationships and a lifetime of experience as harder to capture. He used Warren Buffett as an example, noting that Buffett began learning about business value at age 13 and, over 80 years later, has developed knowledge that was not fully recorded and cannot simply be reproduced.
But Gates drew a sharp line around well-defined work. He said roles such as telesales, telesupport and accounting department tasks are examples of work where AI could be cheaper and better when properly implemented. He also said almost every entry-level job falls into the broad category where AI can substitute at lower cost.
The implication is not only job loss. It is a shift in how careers begin, how companies hire, and how society supports people whose work is affected. Gates said some companies are already hiring fewer entry-level workers, which he called a modest signal.
Data centers are not the center of his warning
Gates pushed back on the idea that opposing data centers is the most effective response to AI risk. His point was that the harms he is describing would not disappear even if every data center in the United States stopped being built, because data centers can be built globally.
For Gates, the more urgent debate is about how to minimize specific harms. That means looking at who can use powerful systems, whether models can be copied without safeguards, how cyber and bio risks are monitored, and whether government has enough AI expertise to respond.
He said government is not as deeply positioned in AI as it was in earlier cutting-edge technologies where it was a major buyer or funder. In his view, that makes public-sector expertise a central issue. He also said collaboration with industry will be needed, especially on cyber risk.
The policy ideas on the table
Gates is not presenting one complete solution. He described his memo as a starting point rather than a finished policy package. Still, he raised several ideas that show where he thinks the debate should go next.
- Monitoring bio-capable models: Gates says models that can make novel molecules should be subject to monitoring and should not be copyable into places where monitoring logic can be removed.
- Human-reserved jobs: Some roles could be set aside for humans by social agreement, with the details varying from one nation to another.
- Robot and token taxes: Gates discussed taxes connected to AI usage that replaces human labor, with government using the money to help people who lose jobs because of AI.
- More government AI expertise: He said the bureaucracy does not need to grow substantially, but government does need deeper understanding of AI.
- International cooperation: Gates said the US should work with China on defining thresholds such as biomonitoring.
He also said he plans to write a bio memo before the end of the year, building on what he knows from the Gates Foundation and his work on pandemics.
Why Gates still sees an upside
Gates is not arguing for abandoning AI. He described real benefits from AI in vaccine and drug innovation, protein-level and cell-level modeling, and tools such as Biomni at Stanford. He also pointed to AI’s ability to help people deal with bureaucracy.
One example he gave is NextLadder, a group spun off from the Gates Foundation. He described scenarios involving low-income families and people dealing with eviction, getting out of jail, bankruptcy, training programs, benefits and other forms of help. In those cases, he sees AI as a potential agent for people who cannot hire many advisors.
That is the tension running through Gates’s argument. AI may help deliver abundance, improve services and widen access to expertise. But he says the path there includes turbulence, and the discussion about guardrails is already late.