Britain’s AI safety institute expands to San Francisco, the home of OpenAI

The British government is taking a significant step in expanding its AI safety efforts by opening a U.S. counterpart to its AI Safety Institute in San Francisco. This move is aimed at solidifying the UK’s position as a global leader in AI and fostering cooperation with the United States in tackling the risks associated with advanced AI systems.

The AI Safety Institute, which currently has a team of 30 in London, will be recruiting a new team of technical staff in the U.S., led by a research director. This expansion will allow the UK to tap into the talent pool in the Bay Area and collaborate with the world’s largest AI labs, ultimately advancing AI safety for the public interest.

San Francisco, as the home of OpenAI and other tech giants, is a natural location for the U.S. counterpart of the AI Safety Institute. The institute was established in November 2023 during a global event in Bletchley Park, England, and has been making progress in evaluating frontier AI models from industry leaders.

Some of the AI models tested by the institute have shown promise in completing cybersecurity challenges, while others struggle with more advanced tasks. Additionally, there are concerns about the vulnerability of these models to “jailbreaks” and the potential for harmful outputs.

The UK government has secured collaboration from companies like OpenAI, DeepMind, and Anthropic to provide access to their AI models for research purposes. This effort comes at a time when the EU is leading the way in AI regulation with the landmark AI Act, which is expected to set a global standard for AI governance.

As the U.S. counterpart of the AI Safety Institute takes shape in San Francisco, the UK continues to assert its leadership in AI safety and looks to strengthen partnerships with key players in the AI industry. This collaboration and cross-border cooperation are essential in addressing the challenges and risks associated with the rapid advancement of artificial intelligence technology.

Back To Top