Anthropic says a US order is forcing an abrupt model shutdown
Anthropic said it will disable its most advanced AI models after the US government ordered the company to suspend access for foreign nationals, according to the supplied report. The action affects the company’s Fable 5 and Mythos 5 models and was presented by Anthropic as an immediate compliance step tied to national security concerns.
The development marks a sharp escalation in the government’s approach to AI controls. For years, US restrictions have focused primarily on chips, semiconductor tools and other infrastructure needed to build advanced AI. This order, as described in the source, moves the line of control closer to the models themselves by restricting who can access them.
What Anthropic says it was told
According to the source text, Anthropic said it received an export control directive ordering it to suspend access to Fable 5 and Mythos 5 for all foreign nationals. The company said it was not given specific details about the national security concern behind the order.
Anthropic said it understands that the government believes there is a way to bypass, or “jailbreak,” at least one of the safeguards intended to prevent Fable 5 from being used to identify software vulnerabilities. In other words, the concern appears to be that a model with strong capabilities could be manipulated into helping users find exploitable weaknesses in software systems.
The company pushed back on the rationale. In the supplied report, Anthropic said it was given only verbal evidence of what it described as a potential narrow, non-universal jailbreak. It argued that such a finding should not justify recalling a commercial model used by a very large number of people.
A flashpoint in AI export policy
The dispute is important because it suggests a new frontier in AI regulation. Export controls have traditionally centered on the physical and technical supply chain: advanced chips, lithography equipment and related tools. Restricting access to AI services themselves is a different and potentially more disruptive mechanism, especially when cloud access allows models to be used globally without shipping hardware.
If access to frontier models can be limited on nationality grounds, AI companies may face a new compliance burden similar to export-screening regimes already common in defense and dual-use industries. The source does not spell out how Anthropic will technically separate affected users, but its statement indicates the company chose a broad shutdown approach rather than attempting a narrower rollout under uncertainty.
Tension between Anthropic and Washington
The order also lands in the middle of an already strained relationship between Anthropic and the US government. The source says the relationship deteriorated earlier this year after Anthropic refused to allow the US military to use its AI models for domestic surveillance and fully autonomous weapons systems. In response, the government placed the company on a supply chain blacklist set to take effect later in the year.
That background matters because it makes the latest directive look like more than an isolated technical disagreement. It arrives amid a broader fight over how AI firms should cooperate with state power, where the boundaries of acceptable defense use should sit, and how much leverage governments should exercise over privately developed models.
The report also notes that signs of easing had recently emerged in disputes between Anthropic and Trump administration officials. The new order cuts against that impression and suggests the underlying conflict remains unresolved.
Why the jailbreak question matters
Anthropic framed the issue as a dispute over how governments should evaluate jailbreak risk. In the supplied account, the company had only days earlier called for greater US oversight of AI, including the ability to block models with unacceptable risks. But it said the Friday action did not follow what it viewed as a principled or evidence-based process.
That distinction is central. AI companies increasingly accept that some level of oversight is inevitable, and some actively support it. The harder question is what thresholds trigger intervention. If a narrow safeguard bypass is enough to halt international access to a major commercial model, then companies may argue that nearly any advanced system could be vulnerable to sudden restrictions. Regulators, on the other hand, may view that very possibility as the reason to move early.
A signal for the wider industry
The reported order is likely to be watched closely by every frontier AI developer. Even without more technical detail, it shows that US authorities may be prepared to act directly against model access when they believe national security is implicated. That creates fresh uncertainty for providers serving global users through a single platform.
For Anthropic, the immediate result is operational and reputational disruption. For the wider industry, the episode may become a precedent. If governments shift from controlling AI hardware to controlling AI capabilities in service form, companies will need stronger screening, clearer deployment tiers and more defensible safety arguments. The supplied report suggests that transition may already be underway.
This article is based on reporting by The Guardian. Read the original article.
Originally published on theguardian.com
