AIAcademy · AIAcademy · 2026-05-16
Read the AI Security Institute (UK)
The institutes did not disappear. They were renamed, in a coordinated-looking but not actually coordinated way, and the rebrand tells you most of what changed and almost everything about the politics.
The UK's AI Safety Institute became the AI Security Institute in February 2025. The word swap was deliberate: "safety" had become culturally loaded — read by some governments as a left-coded alignment-research agenda, read by others as a brake on national-competitiveness AI policy. "Security" reframes the same building, the same staff, and substantially the same eval workstreams around national-security framings that play more comfortably with defence ministries and intelligence services. The technical program — frontier model evaluations, agentic red-teaming, deployment-decision support — barely changed.
The US AISI became CAISI, the Center for AI Standards and Innovation, inside NIST, in June 2025. The renaming was politically heavier — the new administration treated the original AISI as a regulatory-burden vehicle and explicitly resited it as a standards-and-innovation body. The September 2025 announcement that CAISI would work directly with OpenAI and Anthropic on pre-deployment evaluation made clear the core technical mission survived the rebrand. The agreements that the original AISI had signed quietly continued under the new letterhead.
Singapore — never in the renaming wave because they never used "safety" in the institute name — picked up an outsized coordination role around research-priorities setting. The Singapore Consensus on Global AI Safety Research Priorities, drafted at the November 2025 Singapore Conference, became the de-facto agenda for the network's joint workstreams.