The White House announced on Oct 12, 2026 that it is undoing a previous order that stopped the UK AI Safety Institute (AISI) from receiving early copies of cutting‑edge AI models. The change means AISI will again get early access to Anthropic’s Claude Mythos 5.1, Google DeepMind models, and OpenAI’s latest systems. The move follows diplomatic pressure from Britain’s AI minister, Kanishka Narayan.
Why the reversal matters
AISI tests new AI for safety risks before the models are released to the public. Restoring its access helps catch problems early, which could protect critical infrastructure and limit misuse. The White House has not said the policy will stay in place permanently, and it remains unclear how quickly the testing will resume.
Why it matters
For ordinary users, renewed safety testing could mean fewer unexpected harms when new AI tools appear, especially in sectors like energy or finance. It also shows that diplomatic pressure can influence U.S. AI policy, but the long‑term consistency of this access is still unknown.
