Bill Gates Says AI Could Drive Deadly Events

Bill Gates Warns of Mass-Casualty Risks From AI

Artificial intelligence could help people carry out attacks on a scale previously beyond their reach, Bill Gates warned in a Meet the Press interview that aired on Sept. 27. Asked whether AI could end humanity, the Microsoft co-founder described a grave possibility rather than predicting an outcome. Gates stated:

“AI is certainly powerful enough to drive events that, you know, cause a billion deaths.”

His warning about AI-enabled attacks centered on the combination of increasingly capable tools and people intending harm.

Gates identified biological and cyberattacks as immediate concerns. He contended that AI had crossed dangerous thresholds in both areas this year, while the government lacks technical expertise after much of the development took place in the private sector.

Why Gates Calls AI an Evolutionary Event

In the interview, he called the technology an “evolutionary event” and urged policymakers, researchers, voters, and companies to take part in decisions about its use. He described AI as already smarter than people in some respects and predicted rapid gains over the next several years. He argued that AI systems can transfer knowledge without the limits of individual human learning. His August essay on AI’s benefits and harms made a related argument: The choices made now could determine whether the technology reduces inequality or deepens it.

The interview extends concerns Gates had raised before Sunday’s broadcast. In earlier comments on AI oversight, he proposed an international body modeled on the International Atomic Energy Agency and warned about AI-assisted biological attacks.

Gates Calls for Laws Requiring AI Safeguards

Asked whether the industry could regulate itself, Gates replied:

“No one thinks self-regulation is enough.”

He said legislation was needed and that law enforcement and lawmakers should help determine what those protections include. Gates also argued that a kill switch alone would not prevent attacks. A kill switch would allow a company to shut down an AI service, while monitoring would show what the system did and preserve records for investigators. He acknowledged that compliance would add some work for developers but said it would not dramatically slow their progress.

Cooperation across borders would also be necessary, according to Gates. He compared the challenge with Cold War-era nuclear arms negotiations, then described an AI agreement as harder to reach. He said nuclear weapons had unique features, including access to fissile materials. Gates cautioned that treating AI chiefly as a national competition could encourage rapid development without adequate safeguards. At the United Nations on Sept. 22, Secretary-General Antonio Guterres likewise called for a multilateral AI risk management framework, a shared approach to AI risks across countries, with independent oversight.

Gates’ concern about cyberattacks also applies to software that holds and moves money. More than 40 bitcoin and digital asset organizations have asked leading AI developers to give vetted security researchers controlled early access to advanced systems. Their request for stronger defensive access reflects concern that attackers could use those capabilities to find software weaknesses while defenders lack comparable tools.

AI-Enabled Fraud Shows a More Immediate Risk

Consumers already face smaller-scale examples of criminals using AI to make deception more convincing. The FBI has warned that people posing as federal agents use AI-generated videos and fake websites to approach previous fraud victims. The warning about impersonation scams concerns financial theft, a different kind of harm from the mass-casualty events Gates described. Both illustrate his concern about capable tools in the hands of people seeking to misuse them.

For cryptocurrency holders, phishing messages and fake exchanges can also lead to losses when attackers obtain passwords or recovery information. The steps for avoiding crypto fraud include checking website addresses and contacting organizations through their official sites instead of links sent in unsolicited messages.