More Than 800 Figures Urge Halt to Superintelligence
Geoffrey Hinton, Yoshua Bengio, Steve Wozniak and hundreds of other prominent figures call for a ban on superintelligence development until it is controllable and has public backing.
The Future of Life Institute published a declaration on Wednesday calling for a ban on the development of superintelligent systems until there is scientific consensus on their safety and controllability, along with broad public support. The statement has more than 800 signatures from scientists, technologists, religious leaders, politicians and cultural figures.
Signatories include Geoffrey Hinton and Yoshua Bengio, two of the most influential figures in the development of deep learning; Steve Wozniak, Apple’s co-founder; and several Nobel laureates. The list also brings together figures from sharply different political and cultural backgrounds, including Steve Bannon and Prince Harry. That mix is intended to frame concern about advanced AI as an issue that no longer belongs solely to researchers and technology companies.
A Conditional Ban, Not a Fixed Temporary Pause
The declaration is not simply calling for a pause of a few months, nor does it propose a specific timetable. Its demand is more ambitious: halt the race toward superintelligence until there is a sufficient scientific basis to establish that such systems can be developed and governed safely.
The document calls for a ban that should not be lifted until there is scientific consensus on safe, controllable development, along with strong public support. It is a deliberately demanding formula. It is not enough for a company to say it has conducted internal testing, or for a regulator to approve a specific product.
Superintelligence refers to AI that far surpasses humans in virtually all relevant intellectual tasks. It is distinct from artificial general intelligence, or AGI, which is typically described as a system capable of matching human versatility across many tasks. Superintelligence would be a step beyond that and remains hypothetical for now: there is no public evidence that such a system exists.
From Technical Debate to Disagreement Over Who Decides
The call comes as major AI companies compete to build increasingly capable models, agents that carry out multistep tasks and systems designed to automate intellectual work. OpenAI, Google, Anthropic, xAI and Meta have made the development of more general systems one of their goals, although they differ on timelines and definitions.
The signatories question whether a decision of this magnitude should be left to a small number of labs, investors and governments. The issue is not only that a highly capable AI could fail or be used for harmful purposes. Also at stake is who controls infrastructure capable of concentrating economic, military and political power on an unprecedented scale.
The Future of Life Institute is backing the initiative with public-opinion data pointing to a notable gap between the corporate race and social preferences. In the United States, around 5% of respondents would support moving toward superintelligence without regulation. That figure is not equivalent to a political mandate to ban it, but it does weaken the idea that accelerating without limits has clear public acceptance.
Broad Agreement Does Not Resolve the Main Obstacle
The declaration matters because of the number and diversity of its signatories, but it does not create a legal obligation. Turning the appeal into a ban would require agreements among countries with competing strategic interests, as well as verification mechanisms that would be difficult to design.
A practical question also remains: there is no universal technical boundary that can precisely identify when a model moves from being highly capable to heading toward superintelligence. Companies can measure performance on tests, programming ability or autonomy in tasks, but no single metric can certify that a system is safe and controllable over the long term.
Even so, the call shifts the center of the debate. It is no longer limited to asking for more safety evaluations of today’s models. It demands a decision first on whether creating an intelligence that surpasses humans should be a legitimate goal, under what conditions and with what democratic authorization.
