The G7 Summit and the State of Global AI Governance
As G7 leaders convened in Alberta, Canada on June 17, 2025, the summit’s agenda prominently featured pressing geopolitical issues, including developments in the Middle East, Ukraine, trade, and energy security. However, a significant topic was notably absent from discussions: governance for leading artificial intelligence (AI) models and systems. This omission stands out, especially considering the G7’s previous role in pioneering voluntary guidelines for AI governance, exemplified by the Hiroshima AI Process and its Code of Conduct.
The Shift in AI Governance Paradigms
The limited focus on frontier AI safety at the G7 summit reflects a broader transformation in the conception of AI governance. In recent years, there was a marked emphasis on AI safety, with significant initiatives such as the Bletchley Park Declaration from the United Kingdom and the European Union’s AI Act. Additionally, leading AI companies made voluntary frontier safety commitments during discussions in Seoul. Yet, 2025 has ushered in a new paradigm prioritizing AI innovation, sovereignty, and technological competition.
This year alone has seen the withdrawal of the EU’s proposed AI Liability Directive and ongoing deliberations regarding the postponement of parts of the AI Act. The UK and US AI Safety Institutes have undergone branding changes, now known as the UK AI Security Institute and the US Center for AI Standards and Innovation. Furthermore, the US has shelved its Biden-era AI risk management framework and repealed the AI Diffusion Framework, which categorized countries based on their chip-receiving capabilities.
Key Deliverables from the G7 Summit
The G7 summit’s key deliverable, the “Leaders’ statement on AI for prosperity”, primarily emphasizes shared economic opportunities, often using terms like “growth,” “prosperity,” and “competitiveness.” Major initiatives announced include a “GovAI Grand Challenge” aimed at accelerating government adoption of AI and a G7 AI Adoption Roadmap tailored for small and medium-sized enterprises (SMEs). Additionally, a bilateral agreement between Canada and the UK was established, fostering collaboration between their respective AI safety institutes and Canadian AI firm Cohere.
The Challenges of Advanced AI Systems
The retreat from governance discussions may reflect the rapid evolution of AI capabilities. Recent models such as GPT-4.5, Gemini 2.5, and Claude Opus 4 demonstrate reasoning capabilities that were unimaginable just two years prior. Enhanced computational power has empowered these AI systems to tackle complex problems that previously required human-level intelligence. Reports from METR indicate that generalist autonomous AI agents are doubling the lengths of tasks they can undertake approximately every seven months, prompting countries to compete in maximizing the benefits of disruptive technologies.
Simultaneously, the infrastructure supporting AI development is expanding rapidly. Gigawatt-scale data centers, such as the United Arab Emirates’ planned 1GW “Stargate” AI campus in Abu Dhabi, are emerging worldwide. Research from the RAND Corporation warns that AI data centers alone may require an additional 10 GW of power capacity this year, with projections indicating a need for up to 68 GW globally by 2027. Nations lacking advanced AI infrastructure risk becoming digitally dependent on those with established capabilities.
The Call for Multilateral Cooperation
Despite these challenges, prominent figures, including Nobel laureates, CEOs of leading AI companies, and AI researchers, continue to express concerns regarding the potential emergence of Artificial General Intelligence (AGI)—a state where AI systems could match or exceed human cognitive abilities across various domains. This accelerating pace of AI capabilities may account for the shift away from multilateral safety frameworks towards national AI policies, such as the UK’s AI Opportunities Action Plan, the US’s AI Action Plan, and the EU’s AI Continent Action Plan.
However, this pivot towards innovation-first policies often overlooks the synergies among allied nations that could ensure the safe and secure deployment of AI technologies. The same AI advancements that promise transformative breakthroughs in fields like medicine, robotics, and education also pose risks that no single nation can manage independently.
Emerging Security Challenges
The security challenges posed by advanced AI systems are pressing. Model weights that could be exploited by adversaries to undermine years of research and development, alongside massive data centers that are vulnerable to cyber and physical attacks, illustrate the urgent need for coordinated action. Furthermore, AI systems capable of accelerating the misuse of chemical, biological, radiological, and nuclear (CBRN) weapons pose significant threats.
Recent decisions by AI companies, such as Anthropic’s classification of its Claude Opus 4 model under a high internal safety classification due to unexpected capability gains, underscore the challenges faced in managing models that exceed established risk thresholds.
Opportunities for G7 Nations
The G7 nations, which control a significant share of the AI value chain, are ideally positioned to tackle these challenges through collaborative efforts. Unlike other multilateral forums, the G7 encompasses the world’s leading AI developers and adopters in conjunction with established security cooperation mechanisms. Prioritizing the security of AI supply chains and coordinating responses to AI-enabled threats through the AI Safety Institute network can serve the interests of all member nations.
Immediate opportunities for collaboration among G7 nations include joint research on capability evaluation and threat modeling, coordinated disclosure processes for AI companies, sharing intelligence on AI-enabled attacks, and developing export controls. These initiatives not only promise mutual benefits but also enhance competitive advantage by fostering a more secure environment for AI adoption.
As the Canadian G7 summit revealed, ongoing gaps in AI coordination remain. In an era where highly capable AI systems can be trained in one country, deployed in another, and potentially cause harm in a third, purely national approaches to AI governance may falter in addressing cross-border risks.
