World leaders, top AI companies set out plan for safety testing of frontier as first global AI Safety Summit concludes

Countries and companies developing frontier AI have agreed a ground-breaking plan on AI safety testing, as Prime Minister Rishi Sunak brough the world's first AI Safety Summit to a close.

In a statement on testing, governments and AI companies have recognised that both parties have a crucial role to play in testing the next generation of AI models, to ensure AI safety – both before and after models are deployed.

This includes collaborating on testing the next generation of AI models against a range of potentially harmful capabilities, including critical national security, safety and societal harms.

They have agreed governments have a role in seeing that external safety testing of frontier AI models occurs, marking a move away from responsibility for determining the safety of frontier AI models sitting solely with the companies.

Governments also reached a shared ambition to invest in public sector capacity for testing and other safety research; to share outcomes of evaluations with other countries, where relevant, and to work towards developing, in due course, shared standards in this area – laying the groundwork for future international progress on AI safety in years to come.

The statement builds on the Bletchley Declaration agreed by all countries attending on the first day of the AI Safety Summit. It is one of the several significant steps forward on building a global approach to ensuring safe, responsible AI that has been achieved at the Summit, such as the UK’s trailblazing launch of a new AI Safety Institute.

The countries represented at Bletchley have also agreed to support Professor Yoshua Bengio, a Turing Award winning AI academic and member of the UN’s Scientific Advisory Board, to lead the first-ever frontier AI ‘State of the Science’ report. This will provide a scientific assessment of existing research on the risks and capabilities of frontier AI and set out the priority areas for further research to inform future work on AI safety.

The findings of the report will support future AI Safety Summits, plans for which have already been set in motion. The Republic of Korea has agreed to co-host a mini virtual summit on AI in the next 6 months. France will then host the next in-person Summit in a year from now.