From CNBC: 2024-05-20 06:30:03

The British government is expanding its AI Safety Institute to the U.S., aiming to strengthen global cooperation on AI safety. The U.S. counterpart in San Francisco will recruit technical staff to evaluate advanced AI systems. The move showcases British leadership in AI and aims to advance AI safety globally.

Since its establishment in November 2023, the AI Safety Institute has evaluated frontier AI models from leading industry players. Models completed cybersecurity challenges but struggled with more advanced tasks. Some demonstrated PhD-level knowledge in chemistry and biology but remained vulnerable to “jailbreaks” and harmful outputs without oversight.

OpenAI, DeepMind, and Anthropic have agreed to open their AI models to the government for research into risks. The British government’s move comes as other jurisdictions, like the EU, race ahead with AI regulations. The expansion to the U.S. aims to tap into tech talent, engage with AI labs, and advance AI safety for public interest.



Read more at CNBC: Britain expands AI safety institute to San Francisco, home of OpenAI