Trump administration blocks foreign access to Anthropic AI, classifying advanced tech as 'national security asset'

The Trump administration has implemented sweeping export controls, completely restricting foreign access to Anthropic's cutting-edge AI models, 'Mitos' and 'Fable'. This decisive action signals a significant escalation in the U.S. government's strategy to manage advanced artificial intelligence as a national security asset.

The immediate trigger for the ban, enacted on June 12, appears to stem from security concerns voiced by Amazon, a key investor in Anthropic. According to media reports, Amazon researchers successfully prompted 'Fable' to generate information potentially useful for cyberattacks. This discovery was reportedly shared with the U.S. government, prompting demands for Anthropic to rectify the vulnerabilities or halt model distribution. When Anthropic reportedly refused, the Department of Commerce imposed the export controls, leading Anthropic to temporarily suspend all user access to the affected models.

Anthropic has contested the U.S. government's decision, arguing that the issues highlighted by Amazon represent minor vulnerabilities common to other publicly available models, such as OpenAI's GPT-5.5. The company maintains that Amazon's success in 'jailbreaking' 'Fable', completely disabling its safety features, is not definitively proven. Some security researchers reportedly concur with Anthropic's assessment.

Beyond the immediate technical dispute, analysts suggest underlying distrust between the Trump administration and Anthropic may have contributed to the ban. The AI company had previously clashed with the administration over its policy of allowing AI models to be used for "all lawful purposes" without restriction.

Looking ahead, experts anticipate a tightening of U.S. government oversight on high-performance AI, akin to controls on semiconductors and advanced weaponry. Ha Jung-woo, former Senior Secretary for AI Future Planning at the Presidential Office, noted the inevitability of such controls, stating, "Since the release of GPT-3 in the latter half of 2020, I have said that top-tier artificial intelligence would eventually become subject to export controls." He stressed the dual need for global cooperation and the development of independent domestic capabilities to prepare for future contingencies.

“

Since the release of GPT-3 in the latter half of 2020, I have said that top-tier artificial intelligence would eventually become subject to export controls. Such events can be repeated at any time, so while cooperating globally, we must also secure our own capabilities in case of emergency.