Who gets to decide how quickly AI moves?

Artificial intelligence (AI) has stirred up more controversy than most things in the past five years. Ever since the introduction of accessible generative AI in 2022, institutions, universities, students, scholars, activists, cognitive specialists, public policymakers and members of the public have all been divided over the ethics of its use. As of the beginning of this month, even the founders of AI companies have taken the same stance. Dario Amodei, chief executive of Anthropic, released an essay calling on AI companies to slow the development of their most advanced models, as the safeguards currently in place are unable to keep up. Shortly after Amodei’s essay, his position was endorsed by Sam Altman of OpenAI, Elon Musk of xAI, Demis Hassabis of Google DeepMind, and Satya Nadella of Microsoft. In addition, Sam Altman said that OpenAI will not be going public in 2026, as further safety advances are still required. While on the surface this may seem like a development in the collective conscience, the problem is far more multifaceted and layered – leaving the industry and the intended audience divided.

Take a look at the products that each AI giant sells. Amodei, Altman, Musk and Hassabis sell models, while Nadella sells the cloud those models run on. Both groups have advised slowing down. On the other hand, Jensen Huang, founder, president, and CEO of Nvidia, sells the chips that these companies buy before they can build anything. Huang does not advise a slowdown and has instead asserted that AI does not need new legislation because market forces can push companies towards safe innovation without the need for regulation. Mark Zuckerberg, whose company gives its models for free, warned that the proposal could expose companies to greater legal risks.

The common thread between these divisive recommendations is profit. Some companies will benefit from a slowdown, while others will earn greater profits by moving faster. Nvidia’s ability to charge high prices has benefitted from the intense competition between these AI companies, which fear falling behind their rivals. If this competitive pressure decreases, companies may be less willing to pay a premium for Nvidia’s chips. Hence, Huang’s opposition to the proposal may reflect both genuine concerns and Nvidia’s commercial interests.

In the world of AI, antitrust law prevents these companies from holding private meetings to decide what they can and cannot do. This is intended to prevent them from coordinating the pace of development and engaging in unfair competition. However, some of these AI companies have now come forward arguing that they should be allowed to work outside antitrust law for the sake of technological safety. Huang and David Sacks, the White House AI lead, have been critical of this request, as it could reduce competition and increase the influence of these AI companies over their rivals. Sacks also questioned the independence of nonprofit evaluators tasked with evaluating AI safety “independently”. In essence, he asks: how is it fair for one company to inspect another?

Amodei’s proposal would deny China access to the most powerful AI chips and advanced semiconductor manufacturing equipment, and prevent smuggling and remote access to computing capacity abroad. China’s foreign ministry countered the proposal, saying that these warnings are a fearmongering tactic aimed at slowing down China’s technological advancement, echoing the Cold War. Amodei did admit that China poses the toughest predicament for his plan. Despite its alleged advantages, the proposal gives American firms leverage over Chinese ones.

Supporters of AI coordination have compared it with the regulation of banks following the 2008 financial crisis. Basel III was introduced only after a major financial crisis had already occurred. Attempts to regulate AI before harm occurs can represent a genuine effort to prevent future risks, rather than reflecting the commercial interests of AI companies. However, there is an important difference between these two cases. Banking regulations were imposed by external regulators, whereas the proposed AI regulations would involve competing companies coordinating with one another.

Before this debate began, researchers had already issued warnings about the capabilities of AI models. OpenAI slowed down the development of its leading systems after a security incident. Amodei and Altman have both claimed that Anthropic and OpenAI will aim to bring in independent evaluators – a practice not seen among “cartels”. Both concerns can coexist: companies may genuinely fear the risks associated with AI while also pursuing their own commercial interests. The more relevant question remains whether the proposed solution adequately addresses the problem it is intended to solve. An agreement between five major American companies, to be overseen by evaluators they fund, would not apply to international competitors such as those in China and would require an exemption from existing competition law. This solution appears to reflect the structure and interests of the AI industry more closely than the nature of the risk it is intended to address.

Much of the debate surrounding the pace and regulation of AI development has taken place without meaningful participation from other regions, including the Arab world. PricewaterhouseCoopers (PwC) estimates that $31.6 trillion will be invested in AI infrastructure worldwide by 2050. Countries that can provide large amounts of cheap, reliable and low-carbon electricity will be particularly well positioned to attract this investment. Hence, the significant energy, land and investment required for AI infrastructure may provide countries capable of supplying these resources with considerable bargaining power. These states could seek a greater role in determining the conditions under which AI companies operate within their jurisdiction. This could include negotiating who evaluates these AI systems, how the evaluations are to be conducted and who bears legal responsibility when autonomous systems cause harm. This means that the debate goes beyond whether AI development should be slowed to questions of who has the authority to make decisions about its development, and whether other countries should accept decisions made elsewhere.