US Lifts Ban on Anthropic's AI Models, Restoring Global Access

The U.S. government has ended its two-week export ban on Anthropic's latest AI models, Fable 5 and Mythos 5. Anthropic announced that access to these programs would be restored, allowing non-American users to utilize them again. The restriction had initially mandated that only Americans could access the AI models, a rule that extended to Anthropic's own staff. To comply, Anthropic had completely blocked both models, impacting global access.

The AI powering Mythos 5 is known for its ability to detect previously undiscovered software vulnerabilities, sometimes over decades. U.S. authorities and select companies had utilized this capability to address security flaws. A primary concern driving the ban was the potential for such advanced AI to be weaponized for cyberattacks if it fell into the wrong hands. Fable 5, released in mid-June, builds on Mythos technology but specifically blocks its cybersecurity and biotechnology functions. Mythos 5, the non-public, full version, is intended solely for government agencies and specific companies to identify and patch security weaknesses in their systems.

“

We will start restoring access to the programs Mythos 5 and Fable 5 today.

The U.S. restriction had also sparked apprehension in Europe. Following the action against Anthropic, OpenAI, the developer of ChatGPT, limited access to its newest AI model to partners coordinated with the U.S. government. The U.S. government's move was triggered by warnings that safeguards designed to prevent the misuse of models like Fable 5 for discovering software vulnerabilities could be circumvented. Anthropic, however, maintained that no such danger existed.

In a letter explaining the lifting of the ban, U.S. Commerce Secretary Howard Lutnick noted that Anthropic had committed to proactively identifying and fixing security risks within its models. This development follows a period of tension between Anthropic and the U.S. government. The company had previously resisted U.S. pressure to allow its AI models to be used in autonomous weapon systems and mass surveillance within the United States, leading the U.S. Department of Defense to classify Anthropic as a supply chain risk.

“

The company has agreed, among other things, to proactively search for and fix security risks in the models.