On October 12, 2026, the UK AI Security Institute (AISI) published “Towards technical best practice for frontier AI development”, its view of the challenges facing frontier AI developers and the technical practices most urgently needed. The post follows a September 2026 commitment by the Prime Minister to work toward principles and standards for developing the most capable AI systems securely and transparently.
AISI argues that capability progress has outpaced several risk-management and oversight practices, and identifies two newer problems: risk now arises during development itself (training, internal deployment and evaluation), and evaluation methods are struggling to keep pace. It lists eight priority areas: assessing and documenting risk; overseeing AI systems throughout their lifecycle; preserving and evaluating oversight mechanisms; detecting, responding to and reporting agentic AI incidents; sharing information on the speed of AI development, including AI’s role in its own development; enabling independent auditors and evaluators; measuring capabilities in security-relevant domains; and measuring alignment.
The institute says it has independently evaluated more than 55 models before release over three years, and that “Much of how frontier developers manage risk today is voluntary.” It states that “AISI is not seeking to mandate requirements for frontier developers”, but says the practices “could be formalised via international standards development organisations”. It welcomes developer commitments including the White House Accord on Super Intelligence.
Why it matters: the list is effectively the UK’s agenda for what a frontier safety standard should cover. It moves the focus from pre-release testing toward controls during training and internal use, the phase where the year’s agent containment failures happened.
What it does not show: the post asks questions rather than setting requirements, and gives no timeline for any standard. It does not describe the terms of AISI’s own access to unreleased models.