Sources familiar with the matter indicate that Anthropic chose not to send its newest model to the United Kingdom's AI Safety Institute for evaluation prior to its public release. The model, Claude Mythos 5.1, was made available exclusively to a select group of approved institutions within the United States, bypassing the UK safety body entirely.

Citing data from the British government, the report suggests that this decision has stirred unease within Whitehall and raised concerns internally at the institute. Anthropic has opted not to comment. Meanwhile, a spokesperson for the UK Cabinet Office stated that the AI Safety Institute "continues to work closely with industry partners, including Anthropic, to enhance the safety of models."

Where the situation stands

The move marks a notable divergence from earlier collaborations, given that Anthropic had previously engaged with UK regulators on safety protocols. By restricting access to Claude Mythos 5.1 to US-based reviewers, the company has sparked debate over the effectiveness of cross-border oversight in artificial intelligence development. For now, the institute maintains that its partnership with Anthropic remains intact, though the lack of access to the latest iteration has amplified scrutiny from government stakeholders.

Why the focus is on a single model's release

The decision carries weight as it tests the UK's ability to monitor frontier AI systems in real time. Without the opportunity to scrutinize the advanced model, the institute faces gaps in its assessment procedures, potentially limiting its capacity to guide policy on emerging risks. Observers note that the incident underscores the voluntary nature of such agreements, leaving safety regulators reliant on corporate cooperation rather than enforceable mandates.