Anthropic Restores Mythos 5 AccesAnthropic Restores Mythos 5 Access: What the US Government Decision Means for GCC AI Partnerss: What the US Government Decision Means for GCC AI Partners

Anthropic confirmed on June 27, 2026 that it has restored access to Claude Mythos 5 for a select group of US organisations that operate and defend critical national infrastructure, following two weeks of direct engagement with federal authorities that began when the US Commerce Department restricted access to Mythos and Fable AI models on June 12. The company stated it is working with the government to expand Mythos 5 availability further and to make Fable 5 available for general use again, suggesting a phased authorisation process that is expected to take additional weeks.

The restoration matters beyond the US context. For GCC governments and sovereign AI programmes that have built infrastructure strategies explicitly around Anthropic partnerships, the two-week suspension and its aftermath clarifies something that was previously implicit: access to frontier AI models is now a national security policy matter subject to government-level authorisation, and that authorisation can be withdrawn and restored on timelines driven by US federal decision-making rather than by commercial agreements.

What Changed on June 12

The suspension followed months of escalating AI model distillation attacks against Claude, in which adversaries trained weaker models on Claude's outputs at industrial scale to transfer its capabilities without authorised access. Anthropic disclosed to the US Senate Banking Committee that Alibaba-affiliated operators generated 28.8 million fraudulent Claude exchanges through 25,000 fake accounts between April and June 2026, the largest such campaign ever documented against any AI company. The US Commerce Department's response was to restrict Mythos and Fable for foreign nationals globally, a decision that simultaneously cut off international allies and adversaries alike.

The collateral impact on the NSA itself, which lost Mythos access during the dispute, was widely reported. This was particularly striking given that NSA chief Joshua Rudd had testified to Congress that Mythos identified vulnerabilities in almost all US classified systems within hours during Project Glasswing testing, a finding that itself illustrated both why the model is considered a strategic asset and why its unrestricted global access was deemed untenable.

The Trusted Partners Framework

The G7 summit in Evian in June produced discussions around a trusted partners scheme through which allied nations could receive structured access to frontier US AI models. French President Macron stated publicly that he expected progress on this within weeks. The framework, as reported by Reuters, could extend to countries or companies vetted and approved by the US government, allowing them to use Mythos-class models for developing stronger cybersecurity defences.

For the UAE, which formalised its position as a trusted US AI partner through the AI Acceleration Partnership and the Stargate UAE infrastructure programme, and for Saudi Arabia, which has its own deepening technology relationship with Washington, the trusted partners framework represents the most credible path to stable, government-backed access to Mythos 5 and future frontier models rather than exposure to the commercial access volatility the June episode demonstrated.

What This Means for GCC AI Strategy

Three implications follow directly for GCC governments and enterprise AI leaders.

• First, sovereign AI programmes that depend on a single frontier model provider carry concentration risk that goes beyond technical dependency. If that provider's flagship model becomes subject to US national security restrictions, the operational continuity of programmes built on that model is directly affected. Sovereign AI infrastructure strategies that include domestically developed model capability alongside commercial partnerships are more resilient to this risk than those that rely exclusively on imported frontier models.
• Second, the trusted partners framework, if formalised, will create a two-tier global AI market: nations with structured, government-backed access to US frontier models, and nations without. Gulf governments that have invested heavily in US AI relationships are well positioned to be in the first tier, but that positioning will need to be actively maintained through diplomatic and policy engagement, not assumed to follow automatically from infrastructure investments.
• Third, OpenAI's simultaneous release of GPT-5.6 Sol under a similar restricted-access model, previewed first to the US government and a small group of trusted partners, confirms this is becoming the standard deployment pattern for frontier cyber AI models globally. OpenAI's GPT-5.6 Sol is competitive with Mythos Preview on cybersecurity benchmarks, using approximately one-third of the output tokens. The GCC now faces a situation where both of the world's leading offensive and defensive AI cybersecurity models are being deployed under government-mediated access frameworks, not open commercial ones.