Cybersecurity Experts Challenge U.S. Export Ban on Anthropic AI Models

A coalition of 76 prominent cybersecurity experts has issued an open letter urging the U.S. government to reverse its recent export control order on Anthropic’s Fable and Mythos AI models. The government’s decision, which forced Anthropic to suspend global access to these tools, was reportedly triggered by concerns that the models could be jailbroken to perform malicious cyber activities. However, the signatories—including industry veterans from companies like Apple, Meta, and Bugcrowd—argue that these restrictions inadvertently disarm the very professionals tasked with protecting critical infrastructure.

The controversy stems from an unpublished Amazon research paper that allegedly demonstrated a method to bypass Fable’s security guardrails. While the government interpreted these findings as a national security risk, experts like Katie Moussouris argue that the paper merely showcased the model’s ability to perform standard defensive tasks, such as identifying and patching known software vulnerabilities. According to the letter, the ability to automate the 'find, fix, and test' loop is essential for modern security teams, and restricting these capabilities leaves defenders at a disadvantage against adversaries who are rapidly adopting similar AI technologies.

This standoff highlights a growing tension between AI safety regulations and the practical needs of the cybersecurity industry. The signatories contend that the capabilities identified in the Amazon paper are not unique to Anthropic’s models and can already be replicated by other widely available AI systems, including those from OpenAI and international competitors. By limiting access to these advanced tools, the government may be hindering the defensive capabilities of the private sector without effectively mitigating the broader risks posed by AI-driven cyber threats. The experts are now calling for a more nuanced approach that balances security concerns with the necessity of empowering defenders.