TL;DR
The UK and Canada have renewed the AI safety partnership they first signed in May 2024, widening it from research cooperation into three work areas: interoperable AI security standards, information sharing that can run through Five Eyes channels, and technical exchanges between the UK’s AI Security Institute and its Canadian counterpart. The announcement, published on Tuesday 13 October, follows the two prime ministers’ AI commitments at last month’s UN General Assembly. It also arrives in a week when 415 scientists have asked for a new global body that would set containment standards, a reminder that bilateral deals are what currently exist.
What changes
The original Science of AI Safety Partnership paired Canadian research strength with British testing and evaluation. The renewal keeps that division of labour and adds detail. On standards, the two governments will work towards interoperable AI security standards and identify priority areas for common international ones. On information, they will deepen sharing arrangements, using Five Eyes channels where reporting is sensitive, and swap lessons on domestic resilience. On technical cooperation, they plan regular exchanges on frontier capabilities, emerging risks and evaluation methods, and will look at exchanging staff and running coordinated evaluations across the two institutes.
The government frames this as the next step after Prime Minister Burnham’s pledge at the UN to make AI central to Britain’s G20 presidency next year, Prime Minister Carney’s signature on the “Call for Control of Frontier AI Models” on 21 September, and a ministerial meeting that Canada’s Minister Solomon convened three days earlier. Both countries also point to a joint Five Eyes statement in June 2026 urging organisations to act on AI-related cyber risk.
The wider week
The renewal is one of several UK AI security moves this week. Nvidia pledged about £189 million of compute to UK agentic safety research, Washington restored AISI’s pre-release access to frontier models, and AISI set out its priorities for frontier developers, proposing that practices be shared between developers first and formalised through standards bodies as they mature.
Set against that, around 400 scientists from 34 countries, Yoshua Bengio and Geoffrey Hinton among them, used an open letter on Tuesday to urge that a global body set containment standards and keep a register of security incidents. In our view the UK-Canada text is the practical version of that ask: two national institutes agreeing what to evaluate, how, and what to share.
Looking forward
The announcement contains no budget, timetable or named deliverables, and “exploring” staff exchanges is a soft commitment. The measure of it will be whether coordinated evaluations produce joint findings that either government publishes.