AI researcher Timnit Gebru is challenging the industry’s renewed focus on scenarios in which artificial intelligence causes human extinction. In an interview with WIRED, she argues that dramatic future-risk claims can distract policymakers and the public from harms that are already measurable.
The debate intensified after an Anthropic researcher resigned over safety concerns and another staff member publicly estimated a greater-than-10-percent chance that AI could kill everyone within a decade. Those statements prompted arguments over whether the risk is credible, exaggerated or useful to companies seeking influence over regulation.
Gebru’s position is that discussions labeled “safety and alignment” often leave out questions about who controls AI systems and who bears their costs. She points to autonomous weapons as a concrete danger and has long studied discrimination, labor and power in large-scale machine learning. Her criticism does not mean AI systems are harmless; it shifts attention from speculative autonomous superintelligence to decisions people and institutions make with today’s technology.
This is an expert argument, not a new empirical estimate of AI risk, and other researchers sharply disagree with it. Its practical significance is political: regulators choosing which risks to prioritize will reach different policies depending on whether they emphasize hypothetical loss of control or current deployments and accountability.