The Secretary of State for Science, Innovation and Technology, The Rt Hon Michelle Donelan MP and US Commerce Secretary Gina Raimondo signed a Memorandum of Understanding (MOU) on Monday 1 April 2024, which will see the US and UK AI Safety Institutes work together to develop tests for advanced AI models.
Although the two AI Safety Institute announced their cooperation at the AI Safety Summit in Bletchley Park, this memorandum formalises the agreement. This memorandum sets out a plan for collaboration between the two governments, which will lead to:
- Shared frameworks and best practices: This entails aligning scientific approaches in practice.
- Joint testing AI Safety: This involves closely collaborating to iteratively develop robust suites of evaluations for AI models, systems, and agents. The US and UK will conduct joint testing on a publicly accessible model.
- Sharing expertise: This will include partnering on research topics and exchanging expert personnel in practice.
The UK's AI Safety Institute will work together with the US AI Safety Institute to help minimise the risks of AI and harness its potential to benefit society. Michelle Donelan highlighted that the partnership “offers the opportunity for them-the United States government-to lean on us a little bit in the stage where they're establishing and formalizing their institute, because ours is up and running and fully functioning.” The agreement underscores that both the US and UK governments recognise the imperative to act now in partnership on AI Safety to ensure a shared approach to AI safety that can keep pace with this global issue.
This partnership underscores that AI safety is a shared global issue requiring international collaboration to be solved. It aims to establish a common approach to AI safety testing with a unified scientific foundation. As fundamental technical research on AI safety and security progresses, it is reassuring to see both sides of the Atlantic working from a similar base.
With one of the three core ambitions of the AI Safety Institute being to facilitate information exchange this partnership also signifies a significant step forward in the progress of this ambition. Michelle Donelan highlights this as the inception of a 'network of institutes.' Further progress on information sharing will be achieved through the institutes' intention to conduct a joint testing exercise on a publicly available AI model.
For further information on Digital Ethics and AI Safety at techUK, please contact tess.buckley@techuk.org.
Click here for the full press release