Ethereum co-founder Vitalik Buterin has ignited a firestorm by proposing an “AI kill switch” — a mechanism analogous to nuclear plant emergency stops — designed to instantly shut down AI systems if they spiral out of control. The suggestion drew near-instant opposition from several prominent technologists.
Two camps, one deepening divide
Buterin laid out on X a split between two visions. The “AI 2040” camp expects superintelligent AI by 2040 unless drastic measures are taken. The skeptic camp argues humanity can manage risks through coordination, warning that heavy regulation could restrict freedom without real benefit. Buterin admitted uncertainty. “If I were confident superintelligence is coming in 2030 by default, I would be closer to the AI 2040 camp… But I am open to slowing or pausing AI development if the risks become significant,” he wrote.
LeCun fires back: it's an engineering problem, not a switch
Meta's chief AI scientist Yann LeCun countered that AI safety is an engineering challenge solvable through iterative design — just like aviation safety. He expressed skepticism about near-term superintelligence, noting that current large language models are essentially “autocomplete machines.” AI commentator Harry Hawk agreed, predicting future AI systems will be built for reliability, just like aircraft have become safer over time. Buterin responded with the concept of Artificial Superintelligence (ASI) — a hypothetical AI that could perform virtually any task without human intervention.
The AI Futures Project report and its critics
The debate was partly fueled by the “AI 2040: Plan A” report from Daniel Kokotajlo's AI Futures Project, which calls for U.S.-China cooperation to slow progress toward superintelligence, including “mutually assured compute destruction.” AI researcher Richard Ngo dismissed the report as exaggerating the speed of powerful AI emergence and ignoring the political challenges it would create inside nations.
Open-source as a safeguard — or a target?
LeCun argued that the real danger is the concentration of AI development within a handful of powerful firms, and that open-source foundation models are essential for “AI sovereignty.” Author Daniel Jeffries accused “safetyists and hawks” of trying to restrict open source and stifle innovation. Adam Thierer, a senior fellow at the R Street Institute and a member of the U.S. House AI Task Force, called the current U.S. AI legislative process “disorganized and secretive.” He backed “permissionless innovation,” advocating the use of existing laws against harmful applications, test zones, and increased research funding — rather than heavy new restrictions.
No side has convinced the other. Buterin's kill switch proposal has only deepened the already stark fault lines in the AI community.