The United Kingdom is making significant strides in artificial intelligence (AI) safety by expanding its AI Safety Institute internationally, with a new office set to open in San Francisco. This strategic move, announced by the U.K. Technology Secretary Michelle Donelan on May 20, 2024, underscores the U.K.'s commitment to leveraging global expertise and fostering international partnerships in AI safety.

The San Francisco office, slated to open this summer, will allow the U.K. to tap into the Bay Area's renowned tech talent pool and collaborate closely with one of the world's largest AI labs, strategically positioned between London and San Francisco. This expansion aims to bolster relationships with key U.S. stakeholders, reinforcing global efforts to enhance AI safety for the public good.

Michelle Donelan highlighted the significance of this expansion, stating, "It is a pivotal moment in the UK’s ability to study both the risks and potential of AI from a global lens, strengthening our partnership with the US and paving the way for other countries to tap into our expertise as we continue to lead the world on AI safety."

The London branch of the AI Safety Institute, already home to a team of 30, is set to scale further, focusing on risk assessment for cutting-edge AI models. This expansion builds on the momentum from the U.K.'s landmark AI Safety Summit held in London in November 2023, which brought together global leaders and influential voices in the AI industry, including Microsoft President Brad Smith, OpenAI CEO Sam Altman, Google DeepMind CEO Demis Hassabis, and Elon Musk.

In conjunction with the expansion announcement, the U.K. also released a selection of recent results from safety testing conducted on five advanced AI models. The findings, which anonymized the models, provided a snapshot of their capabilities rather than categorizing them as "safe" or "unsafe." Notably, the tests revealed that while some models could handle cybersecurity challenges and possessed PhD-level knowledge in fields like chemistry and biology, all models demonstrated vulnerabilities to basic security breaches and struggled with complex, time-consuming tasks without human supervision.

Ian Hogarth, the chair of the AI Safety Institute, emphasized the importance of these assessments, stating, "AI safety is still a very young and emerging field. These results represent only a small portion of the evaluation approach AISI is developing." This empirical assessment of model capabilities is a critical step in advancing the institute's mission to ensure AI technologies are safe and beneficial for society.

The expansion to San Francisco marks a significant milestone in the U.K.'s leadership in AI safety, reflecting a proactive approach to addressing the global challenges and opportunities presented by artificial intelligence.