The US Just Showed How It Will Kill Its Own AI Lead
On June 12, 2026, the US government issued an export control directive ordering Anthropic to suspend global access to its two most capable models, Fable 5 and Mythos 5, for every foreign national, inc
On June 12, 2026, the US government issued an export control directive ordering Anthropic to suspend global access to its two most capable models, Fable 5 and Mythos 5, for every foreign national, including Anthropic's own employees. The stated reason: a jailbreak that lets users ask the model to read a codebase and identify software vulnerabilities.
Anthropic complied. They also published an unusually blunt public rebuttal, arguing the demonstrated capability is identical to what competing models do openly every day, that no universal jailbreak exists, and that applying this standard industry-wide would halt all frontier model deployments.
This is not a story about one model or one jailbreak. This is the first use of a lever. And the lever now has a track record.
The first restriction is always the least consequential, which is exactly why it is the most important. Once the mechanism exists, the threshold for invoking it drops with each use. What begins as "this specific jailbreak on this specific model" becomes the template for "this capability category" and eventually "anything we deem sensitive."
If the US government can do this now, when AI is still early, it WILL do it again when AI is genuinely powerful. That means the people who need access to transformative tools won't have it. That's the future this directive sketches out.
Here is the detail that transforms this from a policy dispute into a strategic blunder: the majority of researchers at Anthropic, OpenAI, Google DeepMind, and Meta AI are foreign nationals. This is not a fringe detail. It is structural. American AI dominance was built on importing the world's best technical talent and giving them resources no other country could match.
This directive did not just restrict foreign users. It told those researchers they cannot fully use the thing they are building. The government flagged as a national security threat the exact capability those engineers perform as their daily work: autonomous code analysis and repair.
The US used to offer three things no other country could: unmatched compute, unmatched capital, and unmatched freedom to work on hard problems. Two of those still hold. The third just got a question mark stapled to it, while the EU, UAE, and UK are actively courting this exact talent pool.
Imagine if the US had banned Excel in the 1980s.
Excel did not just help finance workers do math faster. It restructured what was possible: new financial instruments, new categories of analysis, entire industries that could not have existed without it. The US captured that economic value because the tool was widely deployed, not merely because it was built in Redmond.
AI is the same compounding dynamic, one order of magnitude larger. The country that deploys it broadly, to researchers, companies, governments, and individuals, captures the productivity gains, the new economic models that form around it, and the institutional knowledge of how to use it. The country that builds it but restricts it exports the upside to whoever fills the vacuum.
Policymakers who internalized the semiconductor playbook (restrict the chip, win the race) have not grappled with the fact that AI value accrues through use, not through possession. Chips are supply-side leverage. AI is demand-side. The logic inverts.
China will never impose an equivalent constraint on its own labs or its own workforce. That is not a prediction; it is a description of how state-directed AI programs operate. Every restriction the US places on its own frontier labs is a constraint China does not share. The question of who "wins" at AI stops being about who built the best model and becomes about who actually put capable AI into the hands of people and institutions worldwide. Deployment reach is influence. Restricted deployment is a vacuum, and vacuums get filled.
Anthropic complied under legal compulsion while publicly contesting the directive. The question now is whether this remains a one-time misunderstanding that gets corrected, or whether it is the opening move in a structural pattern where the US simultaneously claims AI leadership and kneecaps the labs building it.
The answer depends on whether anyone makes the economic case to Congress before the next restriction lands. Because the next one will be easier to issue, harder to contest, and applied to a model that actually matters.
[https://www.anthropic.com/news/fable-mythos-access](https://www.anthropic.com/news/fable-mythos-access)By Eduarda Ferreira