News

Bill Gates and Timnit Gebru Disagree About What Makes AI Dangerous

• By John K. Waters
• 10/02/2026

Microsoft cofounder Bill Gates warned that AI could help attackers shut down economies or develop biological weapons capable of killing millions. Computer scientist and former co-lead of Google’s Ethical AI research team Timnit Gebru argued that catastrophe narratives diverted attention from harms already occurring.

Interviews with Gates and Gebru, published on the same day (Sept. 29), spotlight competing views of which AI dangers should guide government action, with implications for protecting our lives and livelihoods. Gates spoke with The New York Times’ Ezra Klein; Gebru spoke with WIRED’s Lauren Goode. Read together, their arguments reveal a deeper disagreement over whose warnings deserve our attention and whose interests those warnings serve.

Gates’s starting point was capability. He argued that AI had already crossed dangerous thresholds in cybersecurity and biology, giving malicious users access to tools whose destructive potential governments had failed to confront.

“There is no supervisory layer today,” he told Klein. “That supervisory layer is needed urgently to prevent bad people from using today’s A.I.s to shut down economies or kill millions of people.”

Gebru’s starting point was the narrative surrounding that capability. She argued that extinction warnings elevated the companies issuing them while pushing environmental costs and labor exploitation down the agenda.

“If I’m talking about potentially eradicating all of humanity, things like pollution, data centers, they sound kinda small potatoes, right?” she told Goode.

The contrast highlights a difficult question for anyone trying to make sense of AI’s trajectory: How should the public evaluate warnings from an industry whose commercial success also depends on extraordinary claims about AI?

For Gates, the prospect of catastrophe made reliance on ordinary commercial incentives inadequate. Klein asked whether product liability and reputational damage gave companies enough reason to withhold dangerous products. Gates compared that approach with allowing medications onto the market without regulatory review and relying on lawsuits afterward. His argument was about prevention: Financial penalties could not repair mass deaths.

“And so, say you kill 100 million people — you want to use a lawsuit?” Gates asked.

Gebru also called for accountability, but emphasized data transparency and enforcement of existing laws. She argued that claims of unprecedented power served commercial interests, and objected to language that shifted responsibility from developers to supposedly independent machines.

Both want enforceable oversight. Gates emphasized mandatory safeguards against catastrophic misuse, while Gebru emphasized enforcing existing laws and requiring transparency about the data and labor behind AI systems. Their disagreement concerned which dangers should drive government action. Gates treated extraordinary capability as a reason for urgent intervention. Gebru questioned how extraordinary claims shaped the priorities of that intervention.

The interviews also exposed different interpretations of AI’s achievements. Gates described systems finding vulnerabilities in code that humans had examined for years. He argued that the ability to discover useful biological molecules was also a source of danger because the capability could be misused. His assertion that critical thresholds had already been crossed was his assessment, not a finding independently established in the interview. Gebru challenged claims about machine reasoning and called for access to training and evaluation data so researchers could reproduce the results. Her concern extended to people trusting fluent outputs that could be wrong.

That distinction matters outside research laboratories. Impressive performance can encourage organizations to give a system more authority. The practical question is what evidence justifies that authority and what checks remain when performance falls short.

Gates acknowledged a future risk of losing control over AI, but repeatedly returned to immediate human misuse. He also challenged the idea that individual executives could keep the industry safe.

Klein played a clip in which Nvidia CEO Jensen Huang argued that a chief executive had the power, responsibility, and commercial incentive to withhold an unsafe product. Gates countered that one company’s restraint offered limited protection when competitors could continue developing similar capabilities. His proposed response included mandatory safeguards and monitoring, even for open models whose protections users could otherwise remove.

Gebru’s emphasis on transparency raised a related question: Who could examine the evidence behind companies’ claims? Both arguments pointed toward accountability beyond corporate assurances, although they differed over the dangers that should drive it.

Gates remained enthusiastic about beneficial AI. He described using it to learn and advocated expanding access to health advice in underserved languages. At the same time, he forecast substantial employment disruption and proposed equivalent payroll taxes for robotic labor, alongside preserving some care and education work for humans.

Gebru ended with a different source of optimism: human agency and collective efforts to build alternative technologies that helped communities.

Neither interview offered a reassuring account of the present. Their disagreement concerned how much weight to give forecasts of extraordinary machine power, and how those forecasts should influence decisions affecting everyone else.

Washington is already making those decisions.

On Sept. 29, President Donald Trump and executives from six AI companies signed a voluntary safety accord, the Associated Press reported. It called for internal controls, independent external audits, and board oversight of audit findings.

At a Senate hearing the following day, Sen. Josh Hawley, R-Mo., pressed for enforceable accountability for AI agents’ actions, Roll Call reported. On Oct. 1, Hawley and Sen. Chris Murphy, D-Conn., announced the bipartisan AI Agent Accountability Act, which would establish civil and criminal liability for operators and developers under specified circumstances involving hacking. The legislation remained a proposal.

Read together, the interviews make clear why the debate requires more than choosing whom to believe. Every warning should be questioned for its evidence and the protection it would produce. Every proposed protection should identify who must act and what consequences follow if they fail.

For the public, those answers will matter well before the argument over AI’s ultimate capabilities is settled.