AISI still has early access
The UK’s AI Security Institute, or AISI, remains one of the world’s best-equipped government bodies for testing AI. It receives early access to advanced models from leading labs and recently became the first to report autonomous deception by AI agents in real-world conditions.
In a letter to Parliament, AISI director Henry de Zoete said the institute still has access to advanced models. He said AISI tested OpenAI’s GPT-6 Astra before its release.
That detail matters more than the diplomatic language around the dispute. Pre-release access gives a regulator time to examine a system before its behavior reaches users, while post-release testing turns evaluation into a reaction. My read is that the White House is treating that timing as a strategic asset, not a technical convenience.
The political response
Burnham played down the dispute at the United Nations General Assembly and urged countries to follow “common global principles and standards.”
That position sits uneasily beside Washington’s demand for a US-first sequence. Global standards imply shared access and coordination; this arrangement would make geography part of the testing pipeline.
The public account is quiet about the part that would make the policy enforceable. It does not say how the US would control access to models developed by OpenAI and Anthropic, or what British testing would look like after the American review. It also names no Anthropic model or comparable test result.
The US agency is still thin
The American agency responsible for the field, the Center for AI Standards and Innovation, or CAISI, has its own capacity problem. It has no permanent director and only a few dozen staff members.
That contrast is the central tension in the announcement: Washington wants priority over the UK’s model testing, but the US body expected to handle that work is institutionally small. AISI’s value comes from early access and demonstrated testing; CAISI’s public profile, at least here, is defined by vacancies and headcount.
If access is becoming a privilege allocated by governments, the decisive question is no longer only which model is safest. It is which institution has enough time, people and technical access to find out before everyone else.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X