TechFlow News, June 14: According to The Wall Street Journal, U.S. government export controls and access restrictions on Anthropic’s models Fable 5 and Mythos 5 stem partly from cybersecurity research conducted by Amazon and communications between AWS CEO Andy Jassy and the White House.
According to reports, Amazon’s submitted research demonstrated that, through a series of prompt-based tests, researchers could induce Fable 5 to output sensitive information potentially usable in cyberattacks—raising security concerns. Subsequently, Andy Jassy escalated these findings to the U.S. government, prompting the White House to implement further restrictions—including prohibiting foreign users from accessing the model.
Anthropic has denied the government’s characterization of the issue as “jailbreaking,” stating that similar vulnerabilities can occur in other publicly available models, including GPT-5.5. Some security researchers support this view, arguing the issue is better understood as a “prompt injection risk” rather than a severe model breakthrough.
Additionally, former U.S. Department of Commerce official Kate Kolen revealed that the White House’s preexisting policy stance toward Anthropic may have influenced this decision. Anthropic and the White House hold divergent positions on AI safety boundaries—for instance, Anthropic has refused to deploy its models for mass surveillance or lethal autonomous weapons systems. Although tensions eased earlier this year, with both sides expanding cooperation, this incident may reignite friction between them.
