AI systems are becoming more powerful, and quickly. Recent incidents have illustrated emerging risks and there have been calls, including from frontier AI developers themselves, for governments to work together on common standards and information sharing.
AI holds enormous potential for productivity, science and society. And co-operation across borders does not mean slowing down that innovation. It means building the trust that these systems are reliable and secure, so that we can make the most of them.
The question is how governments, businesses and the research community can work together to advance innovation and trust at the same time. There is no need to start from scratch: many of the practical tools and standards already exist at the OECD.
Importantly, these tools were designed to improve interoperability across jurisdictions by creating common definitions, shared reporting frameworks, comparable evidence, and consistent approaches to risk management. That foundation is becoming increasingly valuable as countries seek to co-operate on the opportunities and risks of advanced AI.
Building on shared principles
The starting point is the OECD AI Principles. Adopted in 2019 as the first intergovernmental standards on AI, and updated in 2024, they promote innovative and trustworthy AI that respects human rights and democratic values. Principle 1.4 speaks most directly to the current moment in frontier AI. It calls for AI systems to be robust and safe throughout their lifecycle, with mechanisms to override, repair or decommission them if they behave in undesired ways. None of this is new in spirit: governments have always worked with stakeholders to make powerful technologies safe and reliable. What has changed is the pace of development.
From principles to practice
Beyond principles, the OECD offers practical tools. To help put the Principles into practice, there is the AI Policy Toolkit and Due Diligence Guidelines for Responsible AI. A common framework for reporting AI incidents informs international standardisation efforts and regulatory reporting approaches, and an AI Incidents and Hazards Monitor tracks incidents reported in the media to spot emerging trends. The OECD also hosts the voluntary reporting framework for AI incidents from the G7's Hiroshima AI Process, through which AI developers and deployers can share how they manage risk, building transparency and comparability.
The OECD also brings people together. The Global Partnership on Artificial Intelligence (GPAI) convenes governments, industry, civil society and academic experts to work on practical issues including AI incidents, agentic AI, data governance and privacy, compute and investment. The Global Forum on Technology provides a standing venue for dialogue on the risks and opportunities of emerging technologies more broadly. And the Semiconductor Informal Exchange Network helps governments build a shared understanding of the chip supply chains that AI depends on and make them more resilient.
The questions ahead
These tools and venues are not a finished solution. But they are a foundation for the practical co-operation now needed: building a shared evidence base, learning from incidents, comparing approaches to risk management and identifying where further action is required. The next step is to build on them to tackle harder questions together: How can we strengthen the openness that supports innovation while addressing security and safety concerns? How should incidents involving frontier models be reported, responded to and remediated? And how can smaller democracies without deep technical capabilities participate effectively in the AI transition?
No country can answer these alone. It will take learning from one another and developing practical approaches jointly. This is exactly the kind of co-operation the OECD was built to support: a trusted forum where democracies can work through difficult political and practical issues together.
Why trust matters for growth
Getting this right matters, because trust is what turns AI’s potential into real economic gains, and those gains are far from assured. The OECD estimates that AI could add between 0.2 to 1.3 percentage points to annual labour productivity growth across G7 economies over the next decade, depending on how widely it is adopted. OECD research also shows that firms using AI are more productive than those that are not.
However, uptake remains limited. It more than doubled across OECD countries between 2023 and 2025, but in 2025 just 20% of businesses with ten or more employees were using AI. Adoption is also uneven across sectors: highest in ICT (57%) and professional, scientific and technical activities (37%), and considerably lower in manufacturing (19%) and accommodation and food services (12%). The gap between large and small firms remains significant. AI uptake among small firms has grown quickly, yet in 2025 it still stood at just 17%, one-third the level seen in large businesses.
Closing these gaps means helping businesses and the public sector to adopt AI effectively. That requires investment in broadband, compute and data access, as well as skills. It also means showing businesses, governments and the public that AI systems are reliable and robust, and the results they produce are accurate and trustworthy.
International co-operation cannot be an afterthought. Nor should it be a euphemism for stagnation. It is how trust gets built: the trust to move faster, spread the benefits of AI more widely, and enable economies and societies to thrive with AI.