Skip to main content
Claude Fable 5 switched off across the Gulf after US export order
· 6 min read

Claude Fable 5 switched off across the Gulf after US export order

Anthropic disabled Fable 5 and Mythos 5 worldwide on a US directive. For Gulf states betting on American models, controls have moved from chips to models.

Anthropic has disabled its two most advanced artificial intelligence models worldwide, an action that lands squarely on the Gulf states now spending billions to build their own standing in the technology. Claude Fable 5 and the larger Claude Mythos 5, released on 9 June, were switched off on 12 June after the United States government barred all foreign nationals from using them. Since that bar covers every non-American, in every country, Anthropic said it had no way to comply other than turning both models off for all customers everywhere.

For the United Arab Emirates and Saudi Arabia, the timing is pointed. Both have placed large bets on American frontier models while constructing sovereign capacity at home, from Abu Dhabi's G42 to Saudi Arabia's Humain venture and the Public Investment Fund's push into chips. A directive that reaches every Gulf national, by definition, is a reminder that the most capable tools in that strategy sit behind a switch controlled in Washington.

Advertisement

What Anthropic was told

In its public statement, Anthropic said Commerce Secretary Howard Lutnick wrote to chief executive Dario Amodei at 5:21pm US Eastern on 12 June, designating Fable 5 and Mythos 5 as subject to export controls for all foreign nationals inside or outside the United States, including the firm's own non-citizen employees. The government invoked national security but supplied no detailed account. Anthropic understands the concern to be a demonstrated method of bypassing the model's guardrails, shown by asking it to read a codebase and repair software flaws, with biosecurity also referenced.

Why a single jailbreak closed the whole service

Anthropic says it disagrees that a narrow potential jailbreak should be cause to recall a model used by hundreds of millions, and argues the same test applied industrywide would freeze new releases. The order's reach left no middle path. Amazon Web Services, CNBC reported, asked the company to revoke access in all regions. Crucially, Claude Opus 4.8, Sonnet 4.6 and Haiku 4.5 stayed available, so most live systems in the region kept working while only the two newest models went offline.

From chip controls to model controls

The Gulf has spent a year managing American restrictions on the hardware side. A draft arrangement would send the UAE up to 500,000 advanced Nvidia chips a year, and Washington has tied that access to security conditions, as the Middle East Council on Global Affairs has documented. The Fable 5 suspension marks a shift in kind. Controls have moved from the chips a country can buy to the models a country's residents may use, enforced by citizenship rather than geography. For sovereign programmes that assumed a deployed model was a settled asset once procured, that is a different risk profile. Saudi and Emirati buyers, including state-linked investors who have taken stakes in Anthropic, now have to price the chance that a top model can be made unavailable to their own nationals between one afternoon and the next.

The institutional ties are not abstract. Abu Dhabi's G42 has spent two years restructuring its partnerships to satisfy American security reviews, Saudi Arabia stood up the Humain venture to anchor its own model and data-centre ambitions, and a Saudi technology group recently took a stake in Anthropic itself. Each of those relationships rested on the assumption that closer alignment with American firms would secure dependable access. A directive keyed to citizenship rather than corporate structure cuts across that logic, because it does not matter how a Gulf entity is owned if its engineers and its customers are foreign nationals.

For regional buyers the practical response is procurement design. Contracts that name a single model as the production standard now look fragile, and the firms that fared best this week were those running an abstraction layer able to redirect traffic to Opus 4.8 or a non-American model within minutes. Sovereign programmes are likely to read the suspension as support for funding Arabic-capable systems at home, rather than as a reason to abandon American models that still lead on raw capability.

There is a wider strategic reading too. The Gulf has courted both Washington and Beijing on technology, and American officials have at times pressed regional firms over their China ties. A control that can be applied to a deployed model by nationality gives the United States a faster and finer instrument than chip licensing, one that can be turned on without waiting for hardware to ship. For governments that have framed artificial intelligence as central to diversifying their economies beyond oil, the message is that the supply of frontier capability is now a live variable in the relationship with Washington, not a settled purchase that closer partnership has secured for good.

What we know, and what we do not

The confirmed facts are limited: both models are suspended globally, the instruction came from the US Commerce Department, and Anthropic publicly objects. Much of what Gulf planners need is still missing. The precise security finding has not been disclosed. No end date has been given, and Anthropic presents the suspension as compliance rather than a decision within its control. Whether Fable 5 returns, on what terms, and whether the approach extends to other American developers, all remain open. This is the first known instance of a commercially deployed frontier model being halted by direct US federal action, which is why it is being read closely from Riyadh to Abu Dhabi. It is also the operational face of the broader export-control dispute covered in our separate report on the foreign-national ban.

The Gulf's AI strategy has always carried a quiet assumption that capital and partnership would secure reliable access to the best Western models. This week tested that assumption and found it conditional. The practical takeaway for regional chief information officers is not to abandon American models, which remain the strongest available, but to treat single-vendor dependence as a board-level risk with a real precedent attached. Diversified procurement, Arabic-capable alternatives, and the home-grown capacity that G42 and Humain are building stop being prestige projects and start looking like insurance. The procurement decisions taken across the region in the next quarter, rather than the speeches, will show how seriously that lesson has been absorbed.

AI Terms in This Article 2 terms
alignment

Ensuring AI systems pursue goals that match human intentions and values.

guardrails

Safety constraints built into AI systems to prevent harmful outputs.

Advertisement