On September 9, 2026, multiple reports citing Sina Finance drew a clear line between Whitehall in London and San Francisco: Anthropic refused to grant the UK AI Safety Institute (AISI) testing access to its latest cutting-edge model, Claude Mythos 5.1, prior to its release. Pre-release testing was only opened to qualified US institutions, and the list was not made public. Externally, this was described in briefs as a safety evaluation path designed with a US-centric focus; internally, it directly pierced the narrative of the UK’s painstakingly crafted "AI Safety Governance Center" in recent years—AISI was originally set up to conduct safety assessments and participate in safety standard formulation through early contact with cutting-edge models, yet now it was shut out. As a leading AI lab, Anthropic's decisions on who to grant testing access and which path to take for risk assessments are becoming coveted resources and leverage among regulatory bodies globally. UK national security officials are concerned that this refusal might set a precedent for the industry that would leave AISI without substantial supervisory capability in the face of cutting-edge models. Whitehall and AISI have already shown significant signs of vigilance, yet at the time of the report, both sides remained silent; this silence itself represents a significant lapse in the competition for global AI safety regulatory discourse.

The UK Wants to Be AI Safety Referee but Is Excluded

In recent years, the UK government has repeatedly sent the same signal: in this new race for AI safety governance, London aims to occupy the position of a "global leader." To ensure that this positioning is more than just diplomatic rhetoric, Whitehall has bet on a specific institution—the AISI, positioned as a national AI safety research entity. Its task is not merely academic embellishment, but to conduct systematic safety assessments by "early contact with cutting-edge models" prior to model releases, and then to embed the assessment results into safety standards among the UK and its allies, allowing the UK to wield technical and data leverage at the negotiation table of future rules.

This framework has an implicit premise: leading labs that genuinely possess cutting-edge models must be willing to cooperate and submit the models to institutions like AISI for "off-market reviews" before they are publicly launched. Worldwide, governments are competing for the power to review and influence cutting-edge models; those who can see and assess them first have a greater claim to define what "safety" means. Thus, when leading labs like Anthropic choose to limit pre-release testing to qualified US institutions and deny AISI testing access for Claude Mythos 5.1, the UK’s carefully cultivated role as a "global AI safety referee" suddenly loses its grip—the referee can claim to be a neutral authority, but if it cannot even access pre-release footage and technical details, it can only watch passively from the sidelines as a results-driven competition centered on models unfolds, incapable of truly shaping the outcome of the game.

Claude Mythos Open Only to US Institutions

From the perspective of UK officials, the real sting comes not just from a single rejected email, but from the threshold design that has already taken shape behind the refusal: according to reports from multiple media outlets citing Sina Finance, Anthropic clearly stated that the pre-release testing for the latest model, Claude Mythos 5.1, "is currently only open to qualified US institutions," thereby completely excluding UK national security research institutions like AISI from the testing access list. For Anthropic, this merely constitutes adherence to established entry rules; for AISI, which attempts to formulate safety standards through early contact with cutting-edge models, it means being shut out from the most critical gate before the model is born.

More critically, this is not an isolated technical arrangement, but a safety evaluation pathway drawn by nationality. The testing access for Claude Mythos 5.1 is locked within the qualification review framework of the US, effectively stripping UK regulatory bodies from the early assessment chain of the latest model, forcing them to intervene only after the model is public, or even when risks have already spread. This US-centric safety evaluation system is reshaping the distribution of power: those who can raise questions and impose constraints while the model is still in the lab are the ones who hold the initiative in rule design; meanwhile, AISI, blocked from participation, can only face the foregone conclusion afterward, passively digesting the technical and safety choices already made by foreign labs.

UK Officials Worry About Regulatory Erosion

At a security briefing in London, UK national security officials described Anthropic's refusal as a “dangerous precedent.” Their concerns are straightforward: once leading labs routinely restrict pre-release testing of cutting-edge models to US institutions, AISI will no longer be the default review entity but merely one of many "post-facto commentators." For an institution originally expected to intervene and propose safety constraints before models leave the lab, this means its most critical power—early contact and assessment of cutting-edge models—might be systematically stripped away, effectively hollowing out the entire AI safety regulatory framework that the UK has constructed, leaving only regulatory texts and meeting declarations.

This implicit transfer of power has not yet been documented in any official documents. As of September 9, 2026, Whitehall has not issued any public statements, nor has AISI expressed any positions externally; all emotions are sealed within internal memos and anonymous quotes. The briefings reflect a suppressed vigilance: officials are acutely aware that if they cannot prevent similar arrangements from becoming industry norms, the UK will lose its substantial supervisory capability over cutting-edge models in the competition for global AI safety governance, forced to adjust its regulatory discourse reactively around the established technical pathways and risk boundaries of foreign labs.

AI Safety Evaluation Is Being Hijacked by Geopolitics

Behind the tense atmosphere in Whitehall is a broader struggle: who defines the boundaries of "safety," holds substantial control over cutting-edge models. The brief has pointed out that the global competition around AI safety regulatory frameworks is intensifying, with governments scrambling for review power over leading labs and rule-setting authority. Anthropic's refusal to grant AISI access to test Claude Mythos 5.1 before release, locking pre-release safety evaluations entirely within "qualified US institutions," is interpreted by UK officials as forming a US-centric safety evaluation system, consolidating first-hand information on the real risks of the model and corresponding technical discourse firmly within the US domestic regulatory structure.

This arrangement itself is a form of geopolitical sorting: those permitted to access cutting-edge models early can qualify to play the role of "standard-setters" in global AI safety governance. As Anthropic’s corporate safety evaluation path becomes closely tied with the US regulatory framework, countries like the UK, attempting to establish an international AI safety governance center through AISI, find their space for participation forcibly pushed back. They must wait until models have undergone an initial round of “safety shaping” from a US regulatory perspective before they can conduct limited supplementary reviews of pre-existing realities. The seemingly technical conflict over testing access on September 9, 2026, essentially exposes an evolving pattern: AI safety evaluations are shifting from a professional concern to a geopolitical stratification; those excluded from the first round of evaluations will lose proactive supervisory capabilities over cutting-edge models in future global governance.

What’s Next for the UK and Anthropic

At the moment Anthropic refused to open AISI testing access for Claude Mythos 5.1, the power game between leading AI firms and national regulatory agencies was laid bare: whoever controls the entrance to the first round of safety evaluations will hold a higher starting position in the subsequent global rule-making. As of September 9, 2026, Anthropic has not provided any public remedy or special arrangements to the UK, and the responses from Whitehall and AISI remain at the level of internal concerns and assessments, leaving the tense relationship in a state paused yet ready to restart at any moment. Whether the UK can regain safety evaluation access to cutting-edge models through bilateral diplomacy, coordination with allies, or by making sufficiently persuasive innovations in institutional design is currently unknown, but this is precisely the reality check facing the UK's “AI Safety Governance Center” narrative. For other leading labs globally, this friction has created a clear reference point: in future cross-national model release and testing pathways, balancing corporate autonomy, securing core markets, and responding to various national regulations will not just be a compliance issue, but a strategic decision concerning long-term geopolitical alignments; it will continuously remind all parties that in the competition for first-round safety evaluation rights of cutting-edge models, every refusal and concession could reshape the power dynamics of global AI governance.

Join our community, let’s discuss together and grow stronger!
AiCoin Exclusive Hyperliquid Benefit: https://app.hyperliquid.xyz/join/AICOIN88
AiCoin Exclusive Aster Benefit: https://www.asterdex.com/en/referral/9C50e2
On-chain Telegram Community: https://t.me/AiCoinWhaleData
On-chain Community: https://www.aicoin.com/link/chat?cid=N6OVMor5g
AiCoin On-chain Twitter: https://x.com/aicoinwhaledata

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。