Businesses stand to benefit if agencies in different countries approach the testing of new AI models in a consistent manner, an expert in technology law and contracts has said in welcoming a new partnership on AI safety testing announced by the UK and US governments on Tuesday.
Sarah Cameron of Pinsent Masons said the partnership is “a very positive next step” in international collaboration on AI safety.
The memorandum of understanding (MoU) agreed by the UK and US governments provides for closer ties between the UK AI Safety Institute and its namesake in the US. The agencies will, for example, “develop a shared approach to model evaluations, including the underpinning methodologies, infrastructures and processes” and “perform at least one joint testing exercise on a publicly accessible model”.
Under the agreement, the agencies will also “collaborate on AI safety technical research, to advance international scientific knowledge of frontier AI models and to facilitate sociotechnical policy alignment on AI safety and security”. The partnership could further involve “personnel exchanges”, information sharing, and work on developing wider international collaboration on AI safety and standards for testing.
Cameron said: “The MoU builds on agreements formed at the AI safety summit hosted by the UK government last autumn: the so-called Bletchley declaration last autumn achieved consensus among 29 signatories on, among other things, risks posed by ‘frontier AI’ – a concept that refers to next-generation highly capable general-purpose AI models that at least match if not exceed the levels of capability seen in the most advanced AI models today; and a new AI safety risk testing regime was also agreed between prominent AI developers and government agencies in 10 countries – including the UK, US and the EU.”
“The UK-US MoU is focused specifically on managing the risks that could emerge from the operation of the most advanced AI models in the pipeline, but while the agreement is welcome, businesses would benefit from more international cooperation around issues such as standards and on levels of assurance, for example, to be met in respect of other AI models in operation or under development today. This would help AI developers and users navigate what can be a complex and sometimes conflicting web of requirements that apply across borders – and enable AI that can have a positive impact on the economy and society more widely to thrive,” she said.
In this regard, Cameron welcomed reports that the US and EU are in talks about forming a new partnership for a framework on generative AI.
EU law makers are in the final stages of approving a wide-ranging new AI Act, which would introduce a new risk-based regulatory regime for AI in the EU – including new rules that will affect generative AI systems. In the US, an executive order signed by president Joe Biden last autumn provides for new controls to be applied to generative AI – including the labelling of outputs from generative AI systems.
In the UK, the government does not intend to legislate yet for the use of AI specifically, but instead intends to implement a more flexible system of regulation under which UK regulators would have to have regard to five cross-sector principles pertaining to AI when performing their existing sector-based regulatory functions.
“The more the UK, US and EU can work together – rather than just in distinct EU-US or UK-US initiatives – the better, for businesses,” said Cameron. “Encouragingly, at the Alan Turing Institute AI summit last month, representatives from the UK, US and EU talked collegiately about collaboration and how they all have the same shared mission.”
Cerys Wyn Davies, an expert in AI law at Pinsent Masons, said businesses are increasingly looking for help with rolling out AI-based solutions across jurisdictions.
Wyn Davies said: “Whilst policymakers naturally want to extol the virtues of their approach to AI in their own jurisdictions, the reality is that developers and users of AI want to be able to operate internationally and are looking to policymakers to coalesce around consistent regulation and standards – in the UK, EU and US, especially, as well as elsewhere globally.”
Cameron added: “In respect of frontier AI safety specifically, the mini summit due to take place in South Korea in May could yield some further progress in respect of international cooperation.”
Contact an adviser
