US eases restrictions on Anthropic AI model for selected organizations

The United States government has allowed artificial intelligence company Anthropic to restore access to its advanced Claude Mythos 5 model.

The United States government has allowed artificial intelligence company Anthropic to restore access to its advanced Claude Mythos 5 model for a limited group of trusted American organizations. The decision partially reverses restrictions imposed earlier this month over concerns that the technology could pose national security risks if it reached hostile actors or was used without proper safeguards.

Anthropic announced that more than 100 approved organizations will now be able to use Mythos 5, including several major Fortune 500 companies and institutions responsible for protecting critical infrastructure. The company said it received official approval to quickly restore access for these trusted users while continuing discussions with government officials to broaden availability in the future. It also expressed hope that its Fable 5 model would soon return to general public use.

The restrictions were introduced after the US government ordered Anthropic to suspend access to both Mythos 5 and Fable 5 under new export control measures announced on June 12. Officials argued that highly capable AI systems could be exploited by foreign intelligence services or cybercriminals, particularly in countries viewed as national security concerns such as China and Russia. The Trump administration has taken a stricter approach to overseeing the release of advanced AI models developed by leading companies including Anthropic and OpenAI.

Earlier on Friday, OpenAI also confirmed it had delayed the full public launch of its GPT 5.6 model following a request from the US government. The company said initial access would be limited to a small number of carefully vetted partners whose identities had been shared with federal authorities. The move reflects increasing government involvement in controlling the rollout of powerful AI systems capable of advanced cybersecurity and technical tasks.

Despite the latest approval for selected organizations, the government’s process for deciding who receives access has faced criticism from legal experts and technology observers. John Coleman of the Foundation for Individual Rights and Expression questioned the lack of transparency surrounding the selection process. He argued that it remains unclear why some companies have been approved while others have been excluded, warning that such decisions give excessive authority to the government without sufficient public oversight.

OpenAI Chief Executive Sam Altman also raised concerns about the policy. While he supported extensive safety testing before the release of advanced AI systems, he said the government should not be responsible for deciding which customers are allowed to use the technology.

Security experts have warned that AI models such as Mythos could significantly increase the ability of cybercriminals to identify software weaknesses and launch sophisticated attacks if proper safeguards are bypassed. Banking systems, energy networks and other sectors that depend on complex digital infrastructure are considered especially vulnerable to misuse of such technology.

US Commerce Secretary Howard Lutnick said recent cooperation between Anthropic and government officials had made significant progress in reducing security risks associated with the AI models. Although the exact safeguards were not disclosed, Anthropic previously acknowledged that authorities were concerned about techniques that could bypass safety protections built into its Fable 5 system.

Under the updated rules, approved organizations and their non US employees will no longer require export licences to use Mythos 5. However, licensing requirements will continue to apply to companies that are not included on the government’s approved list. Many of the selected organizations are believed to be part of Anthropic’s Project Glasswing, a partnership involving leading technology companies and research institutions.

The latest decision highlights the growing challenge of balancing national security with technological innovation. While government officials argue that tighter controls are necessary to prevent misuse, industry experts believe prolonged restrictions could slow American innovation and allow global competitors to narrow the gap in advanced artificial intelligence development.