US keeps restricting foreign access to Anthropic's Mythos 5 model

The Trump administration continues to restrict foreign governments, companies, and financial institutions from accessing Anthropic PBC's Mythos 5 AI model, even after lifting export restrictions on the company's other recent models, Bloomberg reports.

Who's been left in limbo

Only a limited group of organizations inside the US can currently access Mythos 5. Foreign governments, companies, and financial institutions that expected to get the model when the broader export rules for Anthropic were eased now find themselves in an uncertain position, according to Bloomberg, which describes their status as "limbo."

• The restriction applies specifically to Mythos 5, not to Anthropic's entire product lineup
• The Trump administration has already lifted export restrictions on the company's other recent models
• Access to Mythos 5 remains limited to a narrow group of organizations in the US
• Affected parties include foreign governments, commercial companies, and financial institutions

Why the US restricts AI models selectively

Anthropic PBC is one of the leading US AI labs. Former OpenAI employees Dario and Daniela Amodei founded it in 2021. The company develops the Claude family of models and is considered one of the key frontier AI players, alongside OpenAI, Google DeepMind, and xAI. For several years, Washington has used export controls as a tool of technological containment: first through restrictions on shipments of advanced Nvidia chips to China, then through tiered rules governing access to AI models themselves based on the recipient's country. Throughout 2025, the Trump administration repeatedly revised rules inherited from the Biden team, easing some exports while keeping others in place.

The specific mention of financial institutions stands out. Banks and investment funds worldwide increasingly rely on AI models for risk analysis, compliance, and algorithmic trading, so restricting access to Mythos 5 in particular directly affects their technology plans, even if Anthropic's other products are already available to them.

How controlling models differs from controlling chips

Export controls on finished AI models work differently from controls on Nvidia chips. Hardware shipments can be blocked at customs and at the manufacturer level, but restricting access to a model means controlling the release of weights, API keys, and cloud access to inference. That's why decisions about which models can be exported and which cannot are often made case by case and may look inconsistent from the outside: authorities weigh not only a model's technical capability but also what it enables, such as the ability to accelerate biological, cyber, or military development.

What it means

A case-by-case rather than blanket approach to AI model export restrictions means that a general easing of policy doesn't guarantee foreign partners predictable access to specific technologies. Companies and governments outside the US will have to navigate each Anthropic model individually rather than rely on the overall policy.