Europe and the United Kingdom are refining their approaches to AI model testing as the U.S. government prepares to establish regulations. U.S. allies have been addressing similar AI safety concerns that the Trump administration now faces. Regardless of the U.S. framework outcome, EU and UK officials indicate that it will only be the beginning of their efforts.
President Trump initiated his second term by discontinuing his predecessor's AI strategy and adopting an anti-regulatory stance. However, as AI models have rapidly advanced, the administration is now developing an oversight framework to determine what should be included. Under Trump's executive order, a voluntary AI framework is expected to be released by August 1. The administration is considering a proposal from OpenAI and Anthropic to create a uniform standard that encompasses all capable models, regardless of their open or closed-source status.
In contrast, the EU has approached AI safety as an ongoing technical and policy initiative for several years, adapting its strategies based on past events. A key lesson from Europe is to begin with specific threat scenarios, using historical data on threats like biological weapons to assess how AI could enhance a bad actor's capabilities and the potential systemic risks to society, according to Chris Canal, CEO and co-founder of AI evaluation firm EquiStamp.
The EU also emphasizes the importance of independent verifications rather than relying solely on companies' assessments. Canal has contributed to shaping the EU and UK regulatory approaches. The EU commits to engaging the scientific community to ground AI risk thresholds in peer-reviewed science before integrating them into policy, encouraging researchers to publish their methodologies. In contrast, the U.S. primarily depends on internal or industry-driven standards, with many technical experts moving to the private sector, particularly in the U.S. and U.K. Governments are starting to converge on cyber risks, but the U.S. and its allies are just beginning to address the broader challenges posed by AI.