In a significant move underscored by cybersecurity concerns, Anthropic has suspended access to its advanced AI models, Fable 5 and Mythos 5, following directives related to export control. This decision was reportedly influenced by cybersecurity research conducted by Amazon and discussions between its CEO, Andy Jassy, and the White House.
The findings from Amazon highlighted potential vulnerabilities that could be exploited through the AI models, raising alarms about their use in cyberattacks. The company demonstrated that by utilizing a series of prompts, Fable 5 could potentially disclose sensitive information that might be leveraged maliciously. Following this revelation, Jassy communicated the findings to governmental officials, prompting the administration to restrict the models’ usage by foreign nationals. This restriction poses a unique challenge as many of Anthropic’s own researchers are foreign-born and now find themselves unable to access their own innovations.
In response to the government’s actions, Anthropic has contested the characterization of the situation, referring to it as a “jailbreak.” The company contends that similar vulnerabilities could be identified using other publicly available models, such as GPT 5.5. This perspective has garnered some support within the cybersecurity community. Katie Moussouris, founder and CEO of LutaSecurity, publicly stated that based on her review of Amazon’s paper, it does not constitute a jailbreak. Additionally, former Commerce Department official Kate Koren suggested that the White House’s prior tensions with Anthropic may have played a role in the decision to limit access to the AI tools.
The relationship between Anthropic and the current administration has been fraught with tension. Previously, the Trump administration had expressed strong disapproval of the company’s stance against utilizing its AI technology for mass surveillance of U.S. citizens or developing lethal autonomous weapon systems. In February, directives from the Trump administration led to federal agencies being instructed to cease the use of Anthropic’s AI, which culminated in Secretary of Defense Pete Hegseth classifying the company as a supply chain risk.
Despite earlier reconciliations that seemed to improve relations and efforts to enhance access to Mythos, the recent developments indicate a resurgence of hostilities between Anthropic and the government. This latest directive introduces new complexities to the ongoing dialogue surrounding AI development, cybersecurity, and regulatory oversight, as both parties brace for what appears to be an impending conflict.
