Artificial intelligence (AI) is a global technology without sufficient international governance. Powerful AI systems operate worldwide, and today’s geopolitical rivalry makes international agreement harder to achieve just as cooperation is becoming more urgent. When national governments act at all, they legislate in different ways and at different speeds. AI has numerous international principles, some national regulators and a growing network devoted to evaluating models. What it lacks is a standing international mechanism for effective, cross-border coordination on governance. We should not wait for a comprehensive grand bargain or a major AI-enabled crisis, including cyberattacks by rogue AI agents. Aligned countries can act now and establish a new platform to pursue practical areas for common ground and to build cooperation. The timing is auspicious, as many AI companies themselves have increased calls for new standards to guide them, and Canadian Prime Minister Mark Carney recently stated:
“There’s a need for coordination. Ultimately, a technology stability board, in our view, along the lines of the Financial Stability Board, would make sense.”
There are transnational and global AI risks that all countries have a shared interest in managing, regardless of broader geostrategic rivalry. These include banning AI control over nuclear arms; stopping the enablement of biological weapons development; restricting autonomous cyber operations; and maintaining effective human control over, and alignment with, powerful frontier models. Addressing these risks should have two core tracks. Track one would feature a bilateral agreement between the United States and China on how to manage the key AI risks, followed by an expanded global framework. Track two would see a coalition of willing countries collaborating to develop the shared understanding and tools needed to implement an agreement for managing global AI risks.
The United States and China made some progress in 2024 on a human-control-over-nuclear-use principle and initiated a formal bilateral track of meetings. High-level bilateral meetings continue this week in the lead-up to a scheduled meeting between US President Donald Trump and Chinese President Xi Jinping on September 24 in Washington, DC, which is expected to address shared risks from AI, including the use of open-weight models. Overall, it appears to be a strategic-risk management approach on select issues rather than a framework for comprehensive AI governance.
The second track could proceed through a coalition of aligned powers moving to establish a new institutional structure — proposed here as the AI Stability Board (AISB) — while encouraging both the United States and China to join from the outset, or over time. The ASIB would not be a new organization, but rather a government-led standing forum for governments, companies and independent experts to build international cooperation on AI. It could develop collaborative approaches to assess advanced AI model capability, share information about serious incidents, and provide a forum in which to learn from each other’s experience on best practices for training and the development of guardrails. It could create the base evidence layer needed for AI governance. Countries would continue to regulate AI independently; there would be value in drawing from comparable testing, common reporting practices and open, trusted assessments. Members could start building good practices across borders without a formal international agreement.
The new AISB would be modelled on the Financial Stability Board (FSB), which was rapidly established after a period of international financial turbulence exposed a coordination problem — and significant ongoing global risks — given fragmented surveillance and regulation. The FSB has effectively brought together existing national and international authorities to assess systemic vulnerabilities, improve coordination, and encourage common standards and best practices — with legally non-binding decisions. There are many parallels to the kinds of coordination issues faced with AI today, and the warnings of the need to act quickly to fill the current gap in governance.
CIGI has long recommended an FSB-style model to address shared global challenges relating to emerging technologies and to respond to the need for a new governance model in the digital era. CIGI extended the model’s application to AI governance and presented the idea at Canada’s 2025 G7 industry, digital and technology ministerial meeting in Montreal. While the institutional model is not new, the political conditions for the launch of an FSB-style body for AI may now be in place. For example, in his address to the European Parliament on September 17, 2026, Carney explicitly proposed that “Canada and Europe can join forces to develop AI safety protocols, coordinate common standards, and strengthen transparency.”
A globally representative coalition with the capability and willingness to cooperate could launch the AISB. Canada, EU countries and others with advanced AI ecosystems, credible regulatory institutions and experience in international convening would be well placed to initiate. While government-led, participants would include major AI developers in the private sector, independent evaluators, standards bodies and relevant international organizations. The coalition should be designed from the outset to expand participation, including from developing countries, rather than be a permanent club of advanced economies. The AISB would need to be supported by a small secretariat, housed and managed by an existing institution, perhaps in a middle-power country such as Canada, Singapore or Switzerland. The host country would be responsible for administration and secretariat services, without turning into a large organization or giving itself additional powers.
Its initial mandate could be focused on developing common evaluation protocols, system documentation, incident learning and assessment frameworks. And there is plenty of existing work and expertise for the board to catalyze. The Organisation for Economic Co-operation and Development, the United Nations and the G7 have all developed principles for managing advanced AI systems. National AI safety institutes, standards bodies and industry have developed credible evaluation methods and techniques for information sharing. Given that many corporate AI leaders are calling for agreed standards for the assessment of new models before public release, they should be ready participants.
An AISB can provide a practical starting point. It would be light enough to establish quickly, narrow enough to generate useful results and open enough to attract wider participation. If it proves successful, it could eventually become a more powerfully mandated body or form the AI component of a broader “Technology Stability Board,” which is perhaps envisaged by Carney.
It does not make sense to wait for a grand conference. A small sponsoring group can agree on a short charter, designate a host institution and establish a small secretariat, and convene the key players. Working groups could quickly be struck on model evaluation, system documentation and incident reporting. Results could be brought to the attention of the multilateral development banks, the International Monetary Fund and the G20, accompanied with invitations for wider participation.
AI governance will ultimately require harder agreements. The largest risks will require national regulation, treaties or direct agreements between the biggest players. A new AI Stability Board would not fulfill all the international coordination needed in AI governance, but it could be a start. It could establish several necessary building blocks, including independent assessment, common testing, transparency and incident learning. It could turn voluntary approaches into durable actions and help establish new norms and practices. We can start with the cooperation that is possible now and use it to build the institutions we will ultimately need.
