On Tuesday, Anthropic released Claude Fable 5, the most powerful artificial intelligence (AI) model ever released to the public. Just three days later, the Trump administration forced it offline by executive fiat, invoking national security to bar its use by any foreign national. The shutdown, the first the US government has imposed on a deployed AI model, lays bare the growing fusion of the capitalist state with the technology monopolies and the ruthless drive to harness AI for war.
Fable 5 is the public version of Mythos, Anthropic’s latest frontier model, which the company first released in early April to a small group of technology and cybersecurity firms, and submitted to the government for review, withholding it from the public as too dangerous given its ability to find and exploit critical software flaws. The company described it as its strongest public system yet in software engineering, data analysis and reasoning, and the first “Mythos-class” model for ordinary users. Its unrestricted twin, Claude Mythos 5, which runs on the same model, remains confined to vetted corporate and government partners through Project Glasswing.
From its release, users found the public model heavily censored. Fable 5 carried classifiers across four declared high-risk areas: cybersecurity, biology, chemistry and the distillation of its capabilities by rivals. When a prompt tripped them, the model refused or silently handed the request to Anthropic’s older Opus 4.8 system. Reviewers found it would not answer basic biology questions about mitochondria or mRNA vaccines, and security researchers said it blocked routine code review. Anthropic called its safeguards “overly conservative” and said they were meant to keep adversaries of the US state from using these more capable models.
This was the model the Trump administration moved to suppress. At 1:15 p.m. on Friday, administration officials called Anthropic and gave it 90 minutes to take Fable 5 and Mythos 5 down, citing a national security threat but providing no details. By 5:21 p.m., Commerce Secretary Howard Lutnick had sent Anthropic CEO Dario Amodei a letter requiring they obtain a special license to distribute the models “to all destinations worldwide” and to share them with non-US citizens, warning that “failure to comply will result in prompt criminal and civil penalties.” Because the order reached every foreign national, including Anthropic’s own foreign-born employees, the company disabled both models worldwide.
The directive followed a frantic mobilization of the national security apparatus. Treasury Secretary Scott Bessent, White House Cyber Director Sean Cairncross and Lutnick met Friday morning to discuss concerns raised two days earlier by Amazon CEO Andy Jassy and at least five other firms.
The stated justification does not withstand scrutiny. The administration claimed it had learned of a method to “jailbreak” Fable 5, that is, to trick the model into ignoring its built-in restrictions, here the safeguards against using it to find software vulnerabilities. Anthropic responded that the technique was narrow, not universal, and that the same capability is available from other models, including OpenAI’s GPT-5.5.
The Commerce Department’s own testing had revealed no significant concerns, and several officials acknowledged that Amazon’s document was misleading, since the same capabilities exist in rival models that face no restrictions. Katie Moussouris, chief executive of Luta Security, who reviewed the underlying research, said it was not a jailbreak at all but a defensive technique to limit misuse. “If national defense is the goal, this is an own goal,” she said.
What is presented as a dispute over safety is in reality a conflict within the ruling class over the terms on which AI is harnessed to the US imperialist war machine. Anthropic is no opponent of that process. Its Claude system, built into Palantir’s Maven targeting platform, generated more than 1,000 targets on the opening day of the US-Israeli war against Iran. Amodei has proclaimed his belief in “the existential importance of using AI to defend the United States” and has said his company has “never raised objections to particular military operations.” The quarrel concerns control, not principle.
The real content of “national security” lies in the preparations of US imperialism for war with China, the central preoccupation of broad sections of the American ruling class. One person familiar with the decision told Semafor that the government acted partly on suspicion that a China-linked group had accessed Mythos. China’s open-weight models are far cheaper but have stayed at least several months behind the American frontier models, above all behind Anthropic’s, which has dominated over the past year, and Washington means to preserve that lead.
These restrictions express the eruption of American militarism across the globe, most recently in the war on Iran, which is a prelude to direct confrontation with China. In every theater of this unfolding global war, AI is being bound ever more tightly to the machinery of war.
The present clash is only the latest stage in a protracted conflict in which Anthropic has repeatedly shown its loyalty to the state. On March 3, the Pentagon designated the company a “supply chain risk to national security” after Amodei refused to grant unrestricted access to Claude, and Trump denounced the firm as a “radical left, woke company.”
Yet the same state that punishes Anthropic depends on it. Even as it pulled the models, the administration was advancing a classified contract for the NSA to use Anthropic’s technology for intelligence analysis and vulnerability detection. Valued at close to $1 trillion after its recent confidential filing to go public, Anthropic will not break with the state on which its business rests. It has called the order disproportionate but pledged to restore access “as soon as possible.”
The episode lays bare the process which the World Socialist Web Site analyzed last week in examining Senator Bernie Sanders’ proposed AI “sovereign wealth fund.” Sanders presents state co-ownership of the AI monopolies as a democratic measure to wrest the technology from the billionaires. But the Fable takedown reveals the same fusion of the capitalist state and the AI corporations, imposed by executive fiat rather than through shares and board seats.
One official described the new arrangement to Axios as “a de-facto licensing regime,” adding that companies “will not screw with the White House.” Another said any future model at the Mythos level or above “would need to go through the administration.” This is the meaning of the executive order Trump signed on June 2, requesting that companies submit their most powerful models for review before release. The state is asserting the power to decide what may be built, what may be deployed and for whom.
The same “national security” pretext drives the global assault on democratic rights. All of the major imperialist powers, from the United States to Europe, Britain, Canada and Australia, are abolishing online anonymity, breaking encryption and placing the population under permanent surveillance, as always in the name of “safety.” AI is the indispensable engine of this apparatus, generating kill lists abroad and tracking immigrants and political opponents at home. The power over Anthropic extends the same drive to bring all communication under the eye of the state.
The significance of this action has been deliberately obscured by the corporate media. The Sunday morning talk shows did not even mention the first government shutdown of a deployed AI model, and the major newspapers buried the story on the front pages of their online publications.
AI is a revolutionary technology, built from the accumulated labor of the international working class, with the potential to eliminate drudgery and raise living standards across the world. Under capitalism, it is being turned into an instrument of war, mass surveillance and the destruction of millions of jobs.
The question of who controls this technology cannot be entrusted to the capitalist government, which is the cockpit of world imperialism, nor left in the hands of the private monopolies. It must become the common property of the international working class, through the expropriation of the technology corporations and their transformation into public utilities under workers’ democratic control.
Tech workers, whose own labor is being turned into instruments of war and repression, must take up this fight through the building of the International Workers Alliance of Rank-and-File Committees (IWA-RFC), uniting with workers across every continent. Freeing this technology from the grip of the oligarchy and its war machine is inseparable from the struggle against capitalism, which demands the independent political mobilization of the working class on a socialist and internationalist program.