The global artificial intelligence industry is failing to meet basic safety standards even as adoption accelerates along sharply asymmetric lines, the United Nations and an independent safety institute warned on Tuesday. The US now controls 75 percent of the computing capacity of the world’s 500 most powerful AI supercomputers, while China holds 15 percent, according to a preliminary report presented at the UN’s Global Dialogue on AI Governance in Geneva. Separately, the US-based Future of Life Institute ranked nine leading AI developers and awarded no company higher than a C+ grade, with Anthropic topping the list and France’s Mistral placing last. The institute found that all nine firms are falling short on combating existential threats, including the pursuit of artificial general intelligence.

The concentration of hardware and model access is rapidly becoming a geopolitical fault line. Washington’s sudden decision in June to block foreign access to some of Anthropic’s most advanced models, later partially reversed, underscored a shift in which access to the systems themselves is now a strategic concern. Viewed from Brasília or Riyadh, sovereignty in AI is increasingly a question of having alternatives: if governments can switch between multiple cutting-edge models, any single restriction loses its leverage. The safety report noted that several companies, including Anthropic, have reversed earlier bans on military use of their technology, with US forces reportedly deploying Anthropic’s systems in operations in Venezuela and Iran before the Pentagon itself briefly barred the company over safety disagreements.

Children have emerged as a central pressure point. UN Secretary-General António Guterres told the Geneva dialogue that no minor should become “a guinea pig for unregulated AI” and proposed three binding principles: mandatory safety demonstrations before exposing minors to AI systems, zero tolerance for AI-generated child sexual abuse material, and a requirement that platforms refer distressed children to human help. The scale is vast: UNICEF data cited at the forum shows at least 20 million children in ten countries already use AI tools, adopting them at more than three times the rate of adults. The Internet Watch Foundation recorded a 26,385 percent single-year increase in AI-generated child abuse videos. Yet a Tsinghua University report released in late June found that outright social media bans for minors, such as Australia’s under-16 prohibition, often backfire by driving children to unregulated spaces and VPNs, with no conclusive evidence linking social media use to psychological disorders. Gulf states, where internet penetration reaches 99 percent in the UAE, are pushing instead for network-level protection that blocks abuse material automatically, shifting responsibility from individual parents to infrastructure.

Alongside the safety push, Guterres announced an Environmental Transparency Initiative requiring big tech firms to disclose carbon emissions, water and land use, and proposed that all data centres run on renewable energy by 2030. He also said he would submit a proposal to the General Assembly for a Global AI Fund to finance infrastructure and access in developing countries. The next factual milestone is the Assembly’s consideration of that fund, which will test whether the asymmetry in AI’s benefits can be addressed before the technology’s risks become entrenched.