Cybersecurity experts push back against US ban on Anthropic AI models

The export control mandate, issued last Friday, forced Anthropic to suspend global access to its most potent AI systems. While the government cited national security concerns, it provided little detail regarding the specific threats posed by the tools. Anthropic suggests the decision may stem from an unpublished Amazon research paper claiming that Fable could be 'jailbroken' to perform unauthorized security tasks.

Signatories of the open letter, including former Facebook security chief Alex Stamos and cryptographer Jon Callas, argue the government’s interpretation of that research is flawed. Katie Moussouris, who reviewed the study, noted that the researchers simply prompted the model to identify and patch known vulnerabilities—a standard, defensive security workflow. Moussouris contends that labeling this functionality as a security bypass is a misunderstanding of how AI assists in software maintenance. The group maintains that the techniques described in the Amazon paper are replicable across numerous other AI systems, including OpenAI’s GPT-5.5 and various Chinese models, making the specific targeting of Anthropic ineffective. The experts are now calling for a more transparent, democratic regulatory process that relies on rigorous scientific vetting rather than broad, restrictive mandates.