Europe can still shape the rules of AI

In recent months, international tech leaders have urged European companies to leverage their engineering strengths to develop AI capabilities using open-source models and proprietary industrial data. Amid growing concerns about AI safety, incidents of AI agents bypassing controls and accessing unauthorized systems have emerged, highlighting the need for improved evaluation before widespread deployment.

While European regulations present challenges, they have also fostered expertise in risk classification and sensitive-data management. Middle powers like Singapore, Gulf countries, and Switzerland possess valuable resources and diplomatic networks that could complement Europe's regulatory and industrial expertise to establish an international framework for assessing AI agents.

By conducting practical tests on agent permissions, task reporting, prompt manipulation resistance, and unauthorized access avoidance, countries can share evaluation methods and independent testing tools. Participation from all stakeholders, including the United States, China, and developing economies, would ensure credibility, transparency, and equal scrutiny of models from every country.

The 2027 Geneva AI Summit presents an opportunity to turn discussions into actionable commitments, leveraging the UN's inclusive political dialogue and technical organizations' role in carrying out implementation.