Fears for the United Kingdom's national security have intensified after it was revealed that U.S. AI company Anthropic, under orders from former President Donald Trump, has excluded Britain's AI Security Institute (AISI) from testing its latest model, Claude Mythos 5.1.
The decision means the AISI, a leading global body for assessing AI risks, will not have the opportunity to vet the new system before its public release, a move that has drawn criticism from across the British political spectrum.
This development follows a June directive from Trump instructing AI companies, including Anthropic, to block foreign nationals from accessing cybersecurity-focused models over national security concerns. Anthropic stated at the time that it would disable its most advanced models for all non-U.S. users.
Politicians have voiced alarm over the UK being potentially sidelined in the rapidly advancing field of artificial intelligence. Al Carns, former armed forces minister, questioned the greatest security threats, suggesting that Silicon Valley companies developing AI pose a greater risk than foreign adversaries. Danny Kruger, of Reform UK, echoed these concerns, writing on X that the exclusion jeopardizes British national security and urged the government to pursue diplomatic options and develop its own secure computing capabilities.
Ben Spencer, shadow minister for AI, emphasized the necessity for the UK to retain its ability to assess and manage risks from advanced AI systems, highlighting the AISI's crucial role. He called for government action to ensure continued access to frontier AI models.
The situation comes amid growing unease within the AI development community itself. Jacob Coxon, a safety researcher at Anthropic, announced his resignation, citing concerns that AI companies are engaging in a "hubristic gamble" with humanity's future by relentlessly pursuing superintelligent technology. He warned that these systems could soon possess superhuman capabilities, capable of hacking, revolutionizing fields overnight, and acquiring significant power.
Evan Hubinger, who leads alignment science at Anthropic, acknowledged Coxon's concerns, stating that there is a genuine belief within the company that AI could pose an existential threat. He admitted that Anthropic does not yet have a clear plan to solve the alignment problem for superintelligence and is not on track to do so, expressing worry about risks arising from recursive self-improvement, which he believes is progressing faster than anticipated.
British AI experts have also expressed concern that the UK risks falling behind in the global AI development race. George Balston of the Alan Turing Institute noted worries about the UK's capability to keep pace with AI advancements and its implications for national sovereignty, warning that a lack of capacity at the frontier could lead to the UK being cut off from key AI capabilities.
In response, a Cabinet Office spokesperson asserted that the UK is a global leader in AI security with a well-resourced institute. They stated that the AISI collaborates with industry partners, including Anthropic, to enhance model safety, citing its recent testing of OpenAI's GPT-6 Astra. The spokesperson added that the UK will continue to test advanced models, build scientific understanding of AI risks, and ensure policy decisions are evidence-based, acknowledging that AI risks transcend national borders.
Meanwhile, legislative efforts are underway in Parliament. Alex Sobel, Labour MP, introduced the Artificial Superintelligence Security Bill, which aims to prohibit the development and operation of superintelligent AI. Darren Jones, a Cabinet Office minister, has called for a new multinational treaty for the regulated and safe development of superintelligence, emphasizing a safety-first approach rather than a ban on innovation.