Anthropic Faces Federal Bans on Advanced Models While Lobbying States for Stricter AI Safety Rules

The artificial intelligence company Anthropic has found itself in the crosshairs of the United States government over its newest and most advanced models. Late on a Friday, the company received an export control directive from the Trump administration ordering it to halt access to its Claude Fable 5 and Mythos 5 models.

This suspension applies to any foreign national, whether inside or outside the United States, including Anthropic's own foreign national employees. The administration cited national security authorities for the directive, which arrived just after SpaceX completed its first day of trading following a record initial public offering.

The federal order came only days after Anthropic Chief Executive Officer Dario Amodei published an essay calling for binding government regulations. Amodei argued that frontier AI models should undergo rigorous technical testing and have their releases blocked if they pose public safety threats, comparing them to commercial airplanes.

A Divided Strategy for State-Level AI Regulation

While facing federal restrictions, Anthropic is actively lobbying state governments to pass increasingly strict AI safety laws. This strategy aims to raise the bar on safety rules state-by-state rather than settling on a single, uniform national standard.

This approach directly opposes the strategy of its primary rival, OpenAI, which supports a unified nationwide framework. OpenAI has engaged in a campaign termed reverse federalism to bypass Congress by passing identical bills across different states.

Cesar Fernandez, Anthropic's head of U.S. state and local government relations, stated that the company is not looking to support the same bill across the country. Instead, Anthropic wants state legislatures to continually raise the safety standards for the most capable AI systems.

This push for state-by-state regulation has drawn criticism from venture capitalists and figures close to the Trump administration. Critics argue that a patchwork of different state laws could stifle technological development and innovation.

Cybersecurity Risks and the Power of Mythos

The federal government's security concerns are closely tied to the sheer capability of Anthropic's latest technology. The company previously acknowledged that its Mythos-level technology was too powerful for a wide public release due to cybersecurity risks.

Mythos has the ability to automatically identify security vulnerabilities in software, which could easily be weaponized if misused. Despite these restrictions, Anthropic has committed up to 15 million dollars to a new cyber defense cohort.

Under this initiative, the company is offering multiple state governments 100,000 dollars in credits to use Claude products to test their cyber defenses. This program allows public agencies to evaluate the technology under controlled conditions.

In a previous submission to Faisal D’Souza at the Office of Science and Technology Policy, Anthropic defined powerful AI as systems matching or exceeding Nobel Prize winners in disciplines like biology and computer science. The company, headquartered at 548 Market Street in San Francisco, had previously released Claude 3.7 Sonnet in February 2025 as the world's most powerful commercially-available system.

Applying Claude to Public Benefits Administration

Beyond security testing, Anthropic is partnering with the civic tech nonprofit Code for America to bring its AI into public administration. The partnership was announced at Code for America's annual summit in Chicago by Chief Executive Officer Amanda Renteria.

The collaboration will deploy Claude to help government caseworkers navigate complex public benefits policies through a tool called the SNAP Policy Navigator. This tool is designed to give caseworkers real-time access to federal, state, and county guidance for the Supplemental Nutrition Assistance Program.

The initiative comes as states face mounting pressure to upgrade their data infrastructure under the federal budget reconciliation legislation, H.R.1. The law requires states to share the administrative costs of SNAP if their payment error rates exceed 6 percent.

By integrating Claude into caseworker workflows, Code for America hopes to reduce administrative burdens, review eligibility documents, and draft plain-language communications for benefit recipients. This represents a practical application of the company's technology even as its frontier models face strict government blockades.

Whether Anthropic can successfully convince states to adopt a patchwork of strict safety rules while its own flagship models remain locked down by federal national security directives is a high-stakes gamble that may ultimately stifle its path to a trillion-dollar valuation.