US Government Lifts Restrictions on Anthropic's Advanced AI Models Following Safety Enhancements

The Trump administration has lifted export restrictions on Anthropic's powerful AI models, allowing the company to resume public access starting today. Howard Lotenick, the US Secretary of Commerce, stated that Anthropic worked closely with the government to address risks related to their AI models, Mythic 5 and Pivotal 5, ensuring compliance and reinforcing American leadership in AI.

Anthropic had suspended access to Mythic 5 and Pivotal 5 on June 12, shortly after their release, due to a US export control order that barred foreign nationals from accessing these models, even within the US. Unable to differentiate between American and foreign users, Anthropic restricted access for all non-US citizens, including some employees involved in development.

Mythic, unveiled in April, features advanced cybersecurity capabilities for scanning software code and identifying vulnerabilities. Due to concerns about misuse by malicious actors, access was limited to around 50 trusted organizations. Anthropic simultaneously developed Pivotal, a version with reduced cybersecurity functions and safeguards against high-risk applications such as biological or chemical weapon design and the creation of other large language models.

Following a report by Amazon researchers who bypassed Pivotal's protections to detect minor security flaws, the US government ordered a block on the model. Anthropic responded by enhancing safety mechanisms that now block 99% of the reported exploit attempts, with only minimal, non-detailed information potentially accessible in rare cases.

According to the Wall Street Journal, the restrictions were lifted after Anthropic's Chief Technology Officer, Tom Brown, engaged in discussions with the Commerce Department and other agencies to demonstrate the model's safety for public use. Anthropic expressed gratitude to its users for their patience and to all collaborators involved in redeploying the models.