Artificial Intelligence
White House seeks U.S. review before British AI tests
Politico said the White House asked OpenAI and Anthropic to wait for U.S. agency review before giving new models to the U.K.'s AI Safety Institute, while Anthropic had already limited one model to U.S. users.
- Politico reported that the White House asked OpenAI and Anthropic to delay sharing new models with the U.K.'s AI Safety Institute until U.S. agencies had reviewed them.
- The article said the request came from the Office of the National Cyber Director.
- Anthropic had already limited access to Claude Mythos 5.1 to organisations in the United States.
- Henry de Zoete said in a letter to Parliament that AISI still had access to frontier models and had tested OpenAI's GPT-6 Astra before release.
- The article said Andy Burnham wanted shared global principles and standards, and that CAISI had no permanent director and only a few dozen staff.
Politico reported that the White House asked OpenAI and Anthropic to delay giving new models to the U.K.'s AI Safety Institute until U.S. agencies had reviewed them. The request came from the Office of the National Cyber Director.
Anthropic had already limited Claude Mythos 5.1 to U.S. organisations, according to the article. Henry de Zoete, who leads AISI, said in a letter to Parliament that the institute still had access to frontier models and had tested OpenAI's GPT-6 Astra before release.
Andy Burnham said at the UN General Assembly that he wanted shared global principles and standards. The article also said the Center for AI Standards and Innovation had no permanent director and only a few dozen staff.
Named in this story
People
- Henry de Zoete
- said AISI still had access to frontier models
- Andy Burnham
- played down the dispute and called for shared principles
Companies
- OpenAI
- was asked to hold back new models from AISI
- Anthropic
- had already restricted one model to U.S. organisations
Governments and agencies
- White House
- asked the companies to wait before sharing models with AISI
- U.K.'s AI Safety Institute (AISI)
- was the institute asked to wait for U.S. review
- Office of the National Cyber Director
- made the request
- Parliament
- received Henry de Zoete's letter
- Center for AI Standards and Innovation (CAISI)
- was said to lack a permanent director and to have only a few dozen staff
Products and systems
- Claude Mythos 5.1
- was limited by Anthropic to U.S. organisations
- GPT-6 Astra
- was tested by AISI before release
How the source tells it
The piece was mostly straight reportage, with brief pressure and prestige shading from a forced-choice framing and a superlative description of AISI.
- pressure and urgency casts the companies as facing a forced choice between two bad outcomes
- prestige and novelty uses a superlative description of AISI and a first-to-report milestone