On 22 June, the Five Eyes intelligence alliance issued an unusually public warning that the most advanced artificial intelligence models are on course to surpass current cybersecurity capabilities within months, not years. The joint statement from the United States, United Kingdom, Canada, Australia and New Zealand said frontier AI would “fundamentally transform both offensive and defensive cyber capabilities” and that risk assumptions were becoming obsolete at a pace that demanded an immediate, whole-of-society response. The advisory urged organisations to integrate AI into security operations, patch faulty software swiftly, and treat cyber risk as a core business responsibility rather than a purely technical problem.

The warning coincided with a high-profile regulatory intervention in Washington. Earlier in June, the US Department of Commerce ordered Anthropic to suspend access for foreign nationals to its two most advanced models, Mythos 5 and Fable 5, citing national security concerns. Anthropic complied, disabling the models for all users worldwide after determining it could not reliably filter by citizenship in real time. The company said it was given only 90 minutes to act and received no detailed explanation of the specific threat. The vulnerability that triggered the order was reportedly flagged to the White House by Amazon, Anthropic’s largest investor, after internal testing suggested the model could assist in cyberattacks. Anthropic has disputed the severity of the finding, arguing it was minor and reproducible in other publicly available models.

The episode has exposed a regulatory vacuum and a sharp dispute over process. A coalition of cybersecurity researchers, entrepreneurs and executives signed an open letter criticising the administration’s opaque and hurried approach, warning that removing the best defensive tools without clear justification was dangerous when adversaries were advancing rapidly. Trump administration officials, including adviser David Sacks, defended the move, asserting that a flaw enabling a cyber weapon was inherently serious. President Trump later stated he no longer considered Anthropic a national security threat, and negotiations are continuing. Viewed from Washington, the intervention sits uneasily alongside an otherwise permissive White House stance that has revoked mandatory AI safety reporting requirements in favour of voluntary frameworks, prioritising innovation to stay ahead of China.

The broader strategic context is an accelerating AI arms race. Industry leaders and Vice President JD Vance have articulated the dilemma: a unilateral slowdown could simply cede advantage to rivals, particularly China, whose open-source models are rapidly closing the capability gap. The Five Eyes statement reflects a growing consensus among Western security agencies that the timeline for AI-enabled cyber threats has compressed so drastically that traditional governance cycles are inadequate. A nascent effort, involving fewer than 50 researchers globally, is developing privacy-preserving verification tools on specialised chips that could one day enable mutual monitoring of AI development, akin to Cold War arms control. For now, no formal regulatory framework exists, but the pressure for coordinated action is mounting, with the Anthropic case serving as a stress test for how governments balance national security, innovation, and the need for transparent risk assessment.