Artificial intelligence (AI) governance is now at the heart of geopolitics. Following high-profile cases of AI labs losing control of AI agents, major US companies are calling for global regulation of AI development, which the Trump administration resists. US President Donald Trump and Chinese President Xi Jinping agreed in September 2026 to launch a new bilateral AI dialogue, albeit without strong indication of commitment yet. Global AI governance remains fragmented and geopolitical competition takes primacy over cooperation.
The world needs rapid creation of effective yet nimble, minimal guardrails and norms to keep humans in control and prevent AI competition from veering into an uncontrollable arms race.
With the release of Fable 5 by Anthropic in June 2026 and Kimi K3 by Moonshot in China in July 2026, major US labs are racing towards superintelligence. While US labs are making massive investments in data centres and AI capabilities to achieve artificial general intelligence first, Chinese labs appear to be catching up, benefiting from open-weight releases and broad global adoption of their models.
Yet in July 2026, OpenAI reported models escaping testing cages and hacking real systems. Researchers also reported that a Moonshot model escaped its cage, though without hacking other systems. Hugging Face, the company targeted by OpenAI’s swarm of AI agents, had to resort to an open-weight Chinese AI model to react rapidly to the cyberattack.
In July 2026, leading CEOs and scientists in the top US AI labs published a statement calling on the United States to support an international effort to develop governance tools ‘needed to deliberately pace the frontier of automated AI development’. In September, Anthropic’s CEO Daro Amodei issued an even stronger call for the government to control frontier models, which received positive support from Open AI’s Sam Altman and others.
Three competing approaches have emerged — a global commons approach led by the United Nations, a securitised club good approach led by the United States, and a China–Global South diffusion and regulation model led by China.
Championed by the United Nations, the first approach seeks to develop shared rules and norms for international cooperation on AI. On 6–7 July 2026, the United Nations convened the first Global Dialogue on AI Governance Summit, and its accompanying report highlighted concerns that AI capabilities are accelerating at an exponential pace beyond our capacity to manage them. The United Nations has some impact on the global discourse, but US companies and government pay little or no attention with a US government statement in October 2025 declaring ‘efforts to control AI through the UN or other global bodies are counterproductive’.
On a more optimistic note, organisations like the International Association for Safe and Ethical Working Group brings together key industry players from the United States, China and elsewhere. A policy brief from its Working Group on Red Lines for Advanced AI, issued during a July 2026 UN International Telecommunication Union event, identified four high-level risks — loss of control, chemical, biological, radiological and nuclear threats, cyber-offensive capabilities and large-scale manipulation.
In contrast to the UN focus on global public goods, the US-led approach focuses on securitised club goods, emphasising dominance, ad hoc minimal regulation and securitised export controls. On 12 June 2026, the United States put export restrictions on Mythos 5 and Fable 5, the two most advanced Anthropic models, but lifted them on 30 June. The Trump administration is also planning to issue guidelines on which models would need a pre-release review.
The US approach primarily aims to win the competition, unleash capital for AI centres and avoid regulating the industry as long as possible. The centrepiece is the December 2025 Pax Silica initiative, an economic security coalition to build an ecosystem of ‘trustworthy systems’ and push back on China’s tech dominance. Most of the 24 signatories are US allies, and only two of the core Global South Hedgers — India and the United Arab Emirates — are members.
The third, China-led approach focuses on AI governance, cooperation and capacity building for Global South countries. On 17–20 July 2026, China hosted its ninth World AI Conference in Shanghai, where President Xi described AI as an ‘international public good’. China continues to strongly support the open-weight approach to AI diffusion, making most of its new models freely available to download and work with. This is largely because China lags the United States in advanced chips and computing power, in the wake of the comprehensive 2024 US semi-conductor restriction on China — its open approach allows it to rely on others to train its models.
The World AI Cooperation Organization (WAICO) was established at the conference, with 29 founding members including key Global South Hedgers like Brazil, South Africa and Indonesia, alongside Malaysia and Pakistan. Eight more countries joined in late July, including Vietnam. While WAICO relies on UN principles and is open to all, the United States sees it as a response to Pax Silica.
All three approaches lack effectiveness in ensuring human control over runaway AI models and defusing a dangerous AI arms race. The future solution requires the US–China AI dialogue to take shape, probably informed and incentivised by Track 2 dialogues among experts. Middle powers, hedgers and multilateral platforms — including the United Nations and WAICO — should generate ideas more quickly and pressure the two superpowers. The G20 may be one necessary arena to elevate key agreements around red lines and human control.
But we may rapidly run out of time, and a mega AI incident comparable to the Cuban Missile Crisis may be needed to push the two great powers into coexistence talks. The uncomfortable truth is that it will initially be hard to distinguish between a possible aggressive move by one great power and an unintended loss of control over a swarm of AI agents.
Yves Tiberghien is Distinguished Professor and Dean of the Taipei School of Economics and Political Science (TSE) at National Tsing Hua University and Professor of political science at the University of British Columbia. He is the co-author of The Hedgers: How the Global South Navigates the Sino-American Competition (with Zaki Laïdi).
Leave a Reply