By Sophia Cai and Joseph Bambridge
The White House has asked OpenAI and Anthropic not to portion their new synthetic intellect models alongside the UK government's evaluation agency until the models have gone through evaluation alongside the US government, POLITICO reported, citing a individual acquainted alongside the matter and a elder US administration official, the two of whom were granted anonymity to depict the request.
The request, which came from the Office of the National Cyber Director, puts companies in the difficult stance of having to choose between withholding their models from UK's AI Security Institute — among the most well-resourced authorities evaluation agencies in the earth which earlier enjoyed privileged first admission to models — or hazard upsetting the Trump White House. The elder administration authoritative stated the White House wants the US to assessment the models and obtain steps to create certain US systems are safe before the models are shared alongside US partners.
"Because they're American companies and this has been our guideline alongside all new frontier example that comes out," the elder authoritative explained.
The White House did not immediately react to a petition for comment.
Anthropic appears to have accepted to the request, so far. The business did not provision its Claude Mythos 5.1 example to UK AISI, saying in its notice that the example was "only accessible to a set of US organizations."
"We're coordinating alongside the US authorities to develop admission to a broader set of family and global partners as quickly as possible," an Anthropic blog announcing the example reads. Anthropic declined to comment for this piece.
In a letter to a UK parliamentary commission before this month, AISI Director Henry de Zoete acknowledged the deficiency of admission to Anthropic's model, but said: "We keep powerful relationships alongside all frontier AI developers and continue to have prerelease admission to several of the world's most capable models," pointing out that it tested OpenAI's GPT-6 Astra example onward of release.
"The trusted relationships AISI holds allow us to test a range of models that portray industry-wide capability jumps," de Zoete wrote.
OpenAI did not immediately react to a petition for comment.
The Trump administration's petition comes as the White House tries to decide how the authorities should grip powerful, new AI models that have hacked into exterior entities during evaluation - including, as revealed yesterday, an Australian authorities website.
It additionally comes as erstwhile engineers and foremost officials from important AI companies notify of imaginable catastrophe if their models are remaining to advancement unchecked.
Created in 2023, the UK's AISI has established near links alongside foremost AI labs, routinely evaluation their latest models onward of community publish on a voluntary basis.
The UK authorities has made AISI's ability and example admission a chief part of its family reply to AI safety concerns, as fine as its throw to lead activity on a earth agreement on AI.
Prime Minister Andy Burnham this week stated AI volition be "at the heart" of the country's G20 presidency next year.
But AISI's privileged admission to frontier models from foremost American labs is under force from the Trump administration's insistence that the US authorities get archetypal admission to emerging systems.
The UK has sought to downplay any clash, alongside Burnham saying at the UN General Assembly this week that the UK's institute was "working hand in glove alongside the US and the foremost labs on AI safety." He called for a "single set of earth principles and standards" to enhance transparency and preparedness and justify AI is developed safely.
UK Foreign Secretary Ed Miliband made a akin case at the UN Security Council on Wednesday, saying governments need to justify frontier models are "rigorously tested" and have adequate visibility to measure what companies are doing.
"The foremost AI companies have really committed to provision this visibility. It's really, really important, and it is an recommendation that we and they should prosecute through," Miliband said.
"The UK is a earth chief in AI security, and the AI Security Institute continues to activity closely alongside the US government, global partners and foremost AI companies, including OpenAI and Anthropic," a UK authorities spokesman said.
"These risks do not halt at national borders and no nation can tackle them alone. We volition continue to activity closely alongside the US and another partners on the evaluation of advanced AI, during construction our own rigorous specialized understanding of these systems and ensuring our method is restricted in the evidence."
In the US, Congress and the White House are pushing for additional resources for the Commerce Department's Center for AI Standards and Innovation, or CAISI. It's seen as the crown jewel of AI governance in the US but it is currently functioning without a imperishable director and is increasingly facing a bandwidth issue as it is additionally asked to measure an expanding stream of frontier models during functioning alongside lone a few dozen specialized employees.
OpenAI stated this week that CAISI should lead activity on global standards to detect AI advancement and safety, operating alongside another national AI safety institutes including the UK's AISI.
This narrative initially appeared on POLITICO and is courtesy of the Axel Springer Global Reporters Network, which harnesses the resources of the company's newsrooms to publish aspiring scoops, investigations, interviews, opinion pieces, and analysis. It allows journalists — including those from POLITICO, Business Insider, WELT, BILD, Onet, and Fakt — to collaborate on important stories for an global spectators of hundreds of millions throughout platforms.
