- On June 12, 2026, the US government ordered Anthropic to suspend global access to its Mythos 5 and Fable 5 models under an export control directive, cutting off European businesses and foreign national employees without warning.
- The European Commission responded by announcing an assessment of “practical consequences” for European users, with officials arguing that partner nations should not face discriminatory treatment under US national security controls.
- The shutdown is accelerating European legislative action: the proposed Cloud and AI Development Act targets reduced dependency on foreign AI infrastructure, backed by an estimated €200 billion investment to triple EU data centre capacity.
The US government cut off European access to two of Anthropic’s most capable AI models overnight, and the fallout is forcing a reckoning over who actually controls access to frontier AI. The June 12, 2026 export control directive, justified by undisclosed national security concerns, disabled Mythos 5 and Fable 5 for all non-US customers, including Anthropic’s own foreign national employees, with no advance notice to affected users.
Anthropic’s Global AI Blackout Ignites Transatlantic Tensions
The immediate trigger, according to reports, was intelligence suggesting Fable 5, the publicly available variant of Mythos 5, could be manipulated to bypass safety controls and exploit software vulnerabilities. Mythos 5 had previously been restricted to a select group of trusted partners due to its capabilities in cybersecurity use cases, making the broader shutdown all the more abrupt for those who had come to depend on it.
The European Commission responded quickly, announcing it was assessing the “practical consequences” for European users and warning that such measures should not discriminate against partner nations. European Commission Executive Vice-President Henna Virkkunen argued that Europe represents an economic opportunity, not a security risk, and called for joint cooperation on governing advanced AI models. The message was direct: Europe does not expect to be treated as an adversary under US export law.
Europe’s Urgent Call for AI Sovereignty Intensifies
The shutdown landed as confirmation of a vulnerability European policymakers had long flagged but struggled to act on: the continent’s critical dependence on foreign AI infrastructure. Access to strategically important models, it turns out, can be revoked at any time by a foreign government’s executive order.
The proposed EU Cloud and AI Development Act (CADA), introduced earlier in June 2026, is now central to that response. The legislation aims to reduce external dependencies by building out homegrown cloud and AI capacity. Alongside CADA, the EU intends to use existing tools, the AI Act, the Cyber Resilience Act and the NIS2 Directive, to manage technology risk on its own terms rather than deferring to foreign states’ security calculations.
The investment ambition is substantial. Officials have set a target to triple EU data centre capacity over the next five to seven years, at an estimated cost of around €200 billion. Germany’s SPRIND “Next Frontier AI” competition, which offers €125 million to fund European AI companies, is one concrete example of the push to establish competitive frontier AI labs within the continent.
Divergent Regulatory Philosophies Collide
The Anthropic incident exposed a deeper structural tension: the US and EU are operating from fundamentally different premises about what AI governance is for.
The US approach, as of early 2026, prioritises innovation velocity and national security control. A December 2025 executive order titled “Ensuring a National Policy Framework for Artificial Intelligence” directed federal agencies to develop standards that could preempt conflicting state requirements, though that centralisation remains contested, with active state laws and anticipated constitutional challenges still unresolved.
The EU’s AI Act takes the opposite approach: a risk-based framework with binding rules centred on fundamental rights. Its extraterritorial reach means US companies running high-risk AI systems that affect EU users must comply, regardless of where those companies are physically located. Key enforcement deadlines for high-risk systems fall in August 2026, though pressure from technology companies and the Trump administration has introduced some uncertainty over timing. For a closer look at how those compliance deadlines are structured, see our coverage of the EU AI Act’s staggered compliance schedule.
The practical effect for global organisations is a compliance environment pulled in two directions at once: US controls that can restrict access unilaterally, and EU rules that require local accountability regardless of where the technology originates.
The Path Forward: Negotiations and Domestic Investment
Europe’s response is running on two tracks. The first is diplomatic. The EU wants to accelerate talks with the US on governing advanced AI models, particularly those with significant cybersecurity applications. The agenda is likely to cover defining what qualifies as “advanced cyber AI,” setting reporting thresholds and establishing joint incident response mechanisms. The EU-US Trade and Technology Council provides an existing forum for exactly this kind of coordination, though its record on producing binding commitments is mixed.
The second track is domestic investment, and the Anthropic blackout has made the case more viscerally than any policy paper could. The €200 billion data centre target and the CADA legislation are both expressions of the same political conclusion: that sovereignty in AI cannot be negotiated from a position of dependency. Whether Europe can close that gap before the next disruption is the harder question. For more analysis on enterprise AI strategy, visit our Enterprise AI section.



