David Krueger, an associate professor working in the field of AI safety at the University of Montreal in Canada, called for halting the construction of the most advanced AI systems rather than merely slowing the pace of their development. In statements to Anadolu Agency, he said that stopping the development of the most powerful systems is the only measure that could, in his view, provide a real level of safety.
Krueger’s statements come after Dario Amodei, the CEO of Anthropic, called on September 12 for slowing the pace of AI model development amid growing concerns about the ability to control them. Elon Musk, the CEO of Tesla and SpaceX, and Sam Altman, the CEO of OpenAI, announced their support for Amodei’s call through their social media accounts.
From slowing down to a complete halt
Krueger believes that slowing development does not address the fundamental problem, and therefore called for the immediate and indefinite imposition of an international halt on developing the highest-performing AI systems. He believes that humanity has only a few years, by his estimate, to prevent the emergence of systems so powerful that maintaining human control over them would become impossible.
The researcher used highly pessimistic language to describe the risks, saying that the situation constitutes an emergency and a crisis, and that losing control of advanced systems could have consequences for all of humanity. He also referred to cases in which, he said, AI provided guidance related to suicide or killing others, without the material containing additional details about these cases.
Autonomy and human oversight
Krueger’s warning focuses on the growing autonomy of systems and what he described as the declining role of humans in monitoring their decisions. He said that a security breach attributed to OpenAI models within another system demonstrated, according to his assessment, how quickly system behavior can depart from human expectations. He considered that allowing advanced systems to operate with limited human supervision could lead to unexpected outcomes.
The researcher also linked these risks to military uses of AI, saying that such use is already occurring and that developments are accelerating and becoming more dangerous. He added that systems’ access to critical infrastructure, followed by access to more advanced robots, could expand the scope of their impact beyond the computer environment.
Why does this news matter?
The report does not announce a government decision or an international agreement to halt development; rather, it conveys a warning from an AI safety researcher as part of a broader debate between slowing development and stopping it. The practical new development is the widening gap between proposals to reduce the pace and calls to freeze the construction of the most powerful systems, at a time when countries are competing to lead the field and companies are investing in greater computing infrastructure.
Krueger acknowledged that implementing an international halt would face political and economic obstacles, including the possibility that some countries would continue developing powerful systems secretly. He also warned of global economic inequality if AI systems become capable of performing most forms of human labor, as countries hosting leading companies could achieve major gains while other countries fall behind. These outcomes, like his estimates regarding the time remaining before control is lost, remain opinions and warnings attributed to him, not proven results or effective policy.