Abstract:
Anthropic refuses to submit its latest model to the UK's Artificial Intelligence Safety Institute (AISI) for testing before the new product is released. The move has raised concerns within the British government that technology companies are following the Trump administration's AI protectionist line.
The $1 trillion company, which is preparing for a major IPO in the coming weeks, launched its latest model, Claude Mythos 5.1, last week, calling it the most capable model in life sciences and cybersecurity.
But this is the first time the model has been open only to accredited US institutions, and the UK's Artificial Intelligence Security Institute (AISI) was unable to gain access. AISI is the world's leading organization in the field of cutting-edge large model evaluation.
Multiple British government officials said that AISI was unable to obtain access rights before the model was released, which has triggered alarm within Whitehall (the British central government) and the institute. AISI is the UK’s most important bargaining chip in the global artificial intelligence race.
One government insider said: "National security officials are concerned about this. They really are not testing Mythos 5.1. There is a lot of anxiety on all sides... They are worried that this will set a precedent and AISI will no longer have access to the latest models in the future."
In recent months, the Trump administration has taken an increasingly stringent stance on the AI industry, introducing extensive export controls and temporarily restricting all foreigners (including employees of U.S. AI companies) from accessing Anthropic's top models.
Sources revealed that some people within Whitehall believe that Anthropic refused to open the new version of Mythos model to AISI at the direction of the US government, but this speculation has not been confirmed. AISI, which was established three years ago, has already obtained the new OpenAI model Astra for evaluation in advance before its official release this month.
Anthropic declined to comment. The company said last week that it was "coordinating with the U.S. government to expand model openness to more domestic and foreign partners as soon as possible."
In June this year, the U.S. government implemented export controls on Fable and Mythos, two models of Anthropic that focus on network security, prohibiting foreigners from using the technology. This incident occurred after the above-mentioned controls were introduced.
The United States discovered that the model had a "jailbreak" vulnerability, which is a method that can bypass the built-in security guardrails of the software, and immediately implemented this control. As a result of the discovery, Anthropic removed its cutting-edge model from global markets for several weeks.
This incident intensified the conflict between the White House and Silicon Valley. Silicon Valley has criticized the U.S. government for adopting temporary emergency regulations on cutting-edge AI.
This move has also attracted opposition from many governments, and relevant countries have accused the United States of punishing allies through bans. The European Commission, the EU's executive arm, warned at the time that security risks were a "shared challenge" and that the measures taken "should not constitute discrimination against partner countries."
British government officials are clearly aware that if the UK is unable to obtain the most cutting-edge AI models, the country will not only face security implications, but the economy may also suffer.
Media reported last month that the Cabinet Office, which is responsible for coordinating government affairs, has ordered an urgent assessment of the risk that ordinary consumers and businesses may no longer be able to use the latest version of AI models.
Anthropic's relationship with the U.S. government has been strained since the beginning of this year. The company had previously been in a stalemate with the U.S. Department of Defense over contract terms, and the U.S. subsequently identified it as a supply chain risk entity.
In recent months, relations between the leading U.S. AI lab and parts of the Trump administration appear to have eased, with both parties working closely on the release of Mythos and Fable models.
The British Cabinet Office said: "The Artificial Intelligence Safety Institute (AISI) continues to work closely with industry partners, including Anthropic, to improve model safety."
Comments