The United States government has allowed AI company Anthropic to restore access to its advanced Claude Mythos 5 AI model for a limited group of trusted organisations, partially reversing restrictions imposed earlier this month over national security concerns.
Anthropic announced that more than 100 approved US organisations, including several Fortune 500 companies and institutions involved in protecting critical infrastructure, will regain access to the cybersecurity-focused AI model. The decision follows discussions between the company and US authorities aimed at addressing potential security risks associated with the technology.
The restrictions were introduced after concerns that highly capable frontier AI models could be exploited by hostile actors or foreign intelligence agencies to conduct sophisticated cyberattacks. Anthropic had suspended access to its Mythos 5 and Fable 5 models following a US government export control order issued on June 12.
According to the company, the latest directive allows trusted organisations defending critical infrastructure to use Mythos 5 without requiring export licences for approved non-US employees. However, access for organisations outside the approved list remains restricted.
The move comes on the same day rival AI developer OpenAI announced it was delaying the full public release of its GPT-5.6 model at the request of US authorities, limiting initial access to a small group of vetted partners.
The government’s selective approval process has sparked criticism from legal experts and technology leaders, who argue that there is insufficient transparency regarding how organisations are chosen. OpenAI Chief Executive Sam Altman also questioned the practice of allowing the government to determine which customers receive access to advanced AI systems.
Anthropic said it will continue working with the US government to broaden access to Mythos 5 and eventually restore public availability of its Fable 5 model. Officials indicated that discussions are ongoing to establish a broader framework for releasing frontier AI models while balancing innovation with national security concerns. The company said it remains committed to deploying advanced AI systems responsibly and securely.