BY MIRABEL ODETA
Anthropic has temporarily suspended access to two of its advanced artificial intelligence models after U.S. officials ordered the company to block foreign nationals from using its technology.
The San Francisco-based AI company announced on June 12 that the restriction applied to foreign nationals both inside and outside the United States, including some Anthropic employees. The decision followed a directive from U.S. officials that raised concerns about the possible misuse of the company’s models.
In response to the order, Anthropic paused access to two of its most advanced systems: Claude Fable 5, a model designed for complex coding and software development projects, and Claude Mythos 5, a model built for cybersecurity and biology research applications.
According to Anthropic, government officials were concerned about a possible method of bypassing the safeguards built into Fable 5, a practice commonly referred to as “jailbreaking.” The company said it was informed of the concern but that the government directive did not provide detailed information about the issue.
Anthropic said it believes the situation resulted from a misunderstanding and is working to restore access as quickly as possible.
The company argued that no AI system can currently guarantee complete protection against all attempts to bypass safety measures. Anthropic said that every major AI provider faces similar challenges because advanced models can be targeted by specific techniques that attempt to produce responses outside their intended safeguards.
The company said the discovery of a limited potential vulnerability should not automatically result in the removal of a commercial AI model that has already been deployed. Anthropic warned that applying such a standard across the artificial intelligence industry could slow or prevent the release of future advanced models from major AI developers.
Before the suspension, Claude Fable 5 had been available to the public, while Claude Mythos 5 was offered only to a small group of approved partners beginning June 9.
Anthropic emphasized that the restriction does not affect its other AI models, including Claude Opus, Claude Sonnet and Claude Haiku, which remain available to users.
The dispute comes as the U.S. government continues to increase its focus on the rapidly expanding artificial intelligence sector, particularly around national security, cybersecurity and the development of advanced technologies.
The move follows recent tensions between Anthropic and U.S. defense officials over the company’s approach to government access and the use of AI systems. Anthropic previously raised concerns that unrestricted access to its technology could lead to applications involving large-scale domestic surveillance or fully autonomous weapons, concerns that defense officials rejected.
The disagreement contributed to increased scrutiny of the company’s relationship with the government, including steps that affected Anthropic’s standing as a technology supplier.
The latest action highlights the growing challenges facing AI companies as governments attempt to balance national security concerns, technological advancement and the need for responsible deployment of increasingly powerful artificial intelligence systems.
Original article
Comments