Anthropic PBC has taken significant steps to comply with a directive from the U.S. government that halts access to its leading artificial intelligence models, specifically Fable 5 and Mythos 5. This unprecedented order comes from the Trump administration, citing national security concerns related to the potential access of these technologies by foreign nationals, both inside and outside the United States.
In a statement, Anthropic confirmed that it has fully disabled access to these AI models for all customers. This action marks a significant escalation in the U.S. government’s approach to AI governance, particularly following historical measures taken by both the Trump and Biden administrations to regulate key technologies like semiconductors and supercomputers. However, this recent restriction is notable for its breadth, impacting software directly, which raises both constitutional and commercial implications.
The trigger for this directive was reportedly the discovery of a “jailbreak” vulnerability in the Fable 5 model, which may allow users to bypass safety protocols. Despite acknowledging the vulnerability, Anthropic contested the rationale behind the government’s decision, asserting that such a narrow finding should not warrant a comprehensive recall of a model that serves millions of users. The company emphasized that applying this standard across the AI industry could effectively halt the deployment of new models, stifling innovation.
Reports suggest that Amazon.com Inc. played a role in these events, as its researchers identified security weaknesses in Anthropic’s technology. Amazon CEO Andy Jassy was allegedly in discussions with U.S. officials regarding these vulnerabilities prior to the implementation of the access restrictions. While Amazon confirmed routine consultations with the government on security issues, specifics of the conversations remain undisclosed.
The government’s sweeping limitations could establish a precedent impacting other major AI developers, such as OpenAI, Google’s Alphabet Inc., and Meta Platforms Inc. Industry leaders, including Nvidia’s CEO Jensen Huang and OpenAI’s CEO Sam Altman, have previously advocated for a strategy of bolstering U.S. AI systems on a global scale rather than restricting access.
This development represents a significant pivot in Anthropic’s relationship with the government, which had previously been improving. Earlier this year, tensions flared when the Pentagon raised concerns over the military applications of Anthropic’s technology, resulting in the company being labeled a potential supply-chain risk. This culminated in U.S. agencies being ordered to phase out its products.
Initially released in April to a narrow audience, Anthropic’s Mythos model was seen as too risky for wider distribution due to its cybersecurity capabilities. Although there were indications that the government’s attitude towards the model was shifting—preparing to allow it in key federal agencies—the sudden directive has reversed these expectations.
The response from Trump’s former AI czar, David Sacks, highlighted the administration’s frustration with Anthropic’s inability to address the vulnerabilities promptly. He remarked that the administration hoped for a quick resolution that would lift export controls and facilitate the model’s return to the market.
At a time when AI developers are racing to deliver cutting-edge technologies, the restriction exacerbates the competitive landscape. Both Anthropic and OpenAI aspire to go public soon, chasing profitability while navigating regulatory complexities.
This incident has prompted calls for a collaborative framework between governments and AI developers to responsibly manage the pace and risks associated with technological advancements. Anthropic, acknowledging the significant implications of its technology, previously advocated for a system that would allow for the option to pause or slow down AI developments that could be hazardous.
The European Union has signaled its ongoing assessment of the situation, emphasizing the need for technological sovereignty in light of the emerging challenges posed by powerful AI models. Concerns have been voiced in the tech community regarding the implications of such sweeping government interventions, with many arguing that it could undermine longstanding technological partnerships.
