In June, the US government issued an order to Anthropic, forcing to disable its most advanced AI models amid concerns over national security. The export controls required the AI firm to restrict access to its Fable 5 and Claude Mythos 5 models to foreign nationals, whether inside or outside the country.

Later in the month, the White House requested OpenAI limit the release of its GPT 5.6 model to a small number of government-approved partners because of its advanced capabilities.

Access to Fable 5 was restored on 1 July following the introduction of a nationality-verification regime, while Mythos 5 remains available only to approved organisations.

Yet even in during the short period Mythos and Fable were disabled, the move caused widespread disruption. On 23 June, legal AI start up Legion filed proceedings, arguing that the restriction unlawfully disrupted its business, because members of its software development team based in Canada lost access to the models.

The episode differs from traditional export-control measures affecting technologies such as semiconductors, says Fiona Phillips, partner and head of Marks & Clerk's AI, cybersecurity and data legal advisory practice. “Those regimes generally establish known restrictions in advance, allowing businesses to structure commercial arrangements and compliance programmes accordingly.”

“By contrast, the Anthropic directive was imposed with little notice and applied to foreign nationals regardless of whether they were physically present in the US, making it a particularly unusual intervention.”

The AI restrictions have shown that the technology many businesses depend on can be swiftly removed without warning. What can UK CISOs do to reduce the impact as AI is increasingly integrated into the business?

Operational Risk

AI tools that can find vulnerabilities at scale and work out ways to exploit them are known as dual use: They can be harnessed by businesses for security and by attackers for nefarious purposes. Given the threats frontier AI can pose – in this case fears centred around the potential for jailbreaking the models – many experts think the US government restrictions are understandable.

Yet the US government’s actions also demonstrate how difficult these controls are to apply in practice, Mayur Upadhyaya, CEO of APIContext says. “Restricting access by nationality sounds straightforward, but global organisations don’t operate that way. Development teams, suppliers and customers are inherently international, so enforcing those restrictions without disrupting legitimate innovation becomes incredibly challenging.”

Indeed, if a critical tool or capability can be restricted, marked up, or removed because of decisions made by a third party or foreign government, it becomes “an operational risk, beyond just a technology issue,” says Ben Lipczynski, director of security and regulatory services at Origina.

With this in mind, the main question this incident has raised is: “What does it mean when a tool your business depends on can be switched off overnight?” asks Chris Newton-Smith, CEO, IO.

The risk is substantial partly because the AI market is “very concentrated,” he tells SC Media UK. This means a small number of providers account for the models that many businesses now rely on. “These models are often deeply embedded in how firms work on a daily basis. Such dependency has rarely been stress-tested because, until recently, there has been little reason to do so.”

In a world where access can change because of export controls or geopolitical events, organisations now need to treat foundation models as external dependencies in the same way they think about cloud providers or identity platforms, says Upadhyaya. “That creates questions around sovereignty, resilience and operational control that many organisations haven’t yet considered.”

Future Of Frontier AI

With the US continuing to control AI, the one thing to be sure of is uncertainty. It also comes as the UK National Cyber Security Centre launches its Cyber Shield initiative, a move to use frontier AI in national defence via red and blue teaming.

Security leaders globally are now wary to depend on AI-as-a-service models that “can be launched with much fanfare and then pulled from the global market,” says Andrew Bolster, senior R&D manager at Black Duck.

At the same time, there is increasing global investment in sovereign AI, such as the EU’s Tech Sovereignty Package, he points out. Meanwhile, Gulf States are collaborating on an aligned strategy to use open models such as Mistral and Chinese derived technologies from Alibaba. “In the meantime, the wider open-weight model ecosystem has been nipping at the heels of the professed benchmarks, with GLM-5.2 from Chinese Z.ai.”

Taking this into account, Lipczynski predicts AI access will continue to fragment along geopolitical lines. “The US certainly won't be the last government to assert control over frontier AI capability, and enterprises will increasingly have to factor that into procurement decisions the same way they factor in data residency or regulatory jurisdiction.”

For now, CISOs need to start thinking of AI models as “critical operational dependencies,” says Upadhyaya. He describes a recent conversation he had with the UK general manager of one of the major hyperscalers: “We discussed how many organisations experienced silent failures when Anthropic’s models changed over the weekend. Automations had defaulted to the latest model, and when availability changed, there was no graceful degradation. Workflows simply stopped executing as expected. That’s the operational challenge: Machines fail silently, and at machine speed.”

Rather than asking whether an AI model is available, organisations need to understand which business workflows depend on it, what happens if that dependency changes, whether they have failover to alternative models, and whether they’re continuously verifying that critical transactions are still executing correctly, says Upadhyaya. “If the UK is encouraging AI adoption as a driver of economic growth, resilience around these dependencies needs to become part of the conversation.”

Mature AI governance frameworks are the key to unlocking visibility,  says Newton-Smith. “This will mean organisations know what models they are using, where their dependencies exist, what data is being processed, and which operations will be affected if access was interrupted. In addition, policy structures will help guide employees through the kind of disruption the Anthropic situation has created.”

Kate O'Flaherty Cybersecurity and privacy journalist