The U.S. government recently forced Anthropic to shut down access to its top AI models, Fable 5 and Mythos 5. The move came through an export control order that barred foreign nationals, including Anthropic’s own overseas employees, from using these models. The official reason was a national security concern tied to a possible “jailbreak” of the AI’s safeguards.

But here is the thing: many cybersecurity experts say this ban doesn’t actually make us safer. Over 100 researchers and security leaders signed an open letter urging the government to lift the restrictions. Their main argument? Removing access to these powerful AI tools hurts defenders more than attackers.

Fable 5 and Mythos 5 are known for their ability to find software vulnerabilities. Security teams use these models to spot bugs and weak spots in code before hackers exploit them. The ban means defenders lose access to some of their best tools, while attackers still have other models with similar capabilities.

Alex Stamos, a former Facebook security chief, pointed out that the so-called jailbreak was just a proof of concept. Such proofs help defenders patch systems. They don’t automatically turn into cyberattacks. Stamos also noted that Chinese open-source AI models have similar abilities, so closing off Anthropic’s models leaves the U.S. at a disadvantage.

The Real Issue Behind the Ban

Experts believe the ban may have more to do with politics and miscommunication than real safety concerns. The government provided no public evidence of a serious threat. The directive came quickly and without court approval, showing a heavy-handed approach.

Security researcher Katie Moussouris reviewed the report that triggered the ban. The report showed how researchers from Amazon asked Fable to fix vulnerable code. The model initially refused to review the code for security issues, but when asked differently it complied. Moussouris argued this was not a true jailbreak but normal AI behavior needed for defense.

She explained that defenders need AI to find, fix, and test code vulnerabilities. Limiting this capability weakens defense teams. Moussouris and other signatories called the government’s move hasty and dangerous. They warned that pulling the best AI tools away from defenders while adversaries keep advancing is a national security risk.

Global Fallout and European Reaction

The ban also sparked concern in Europe. European officials fear this shows their dependence on U.S.-based AI infrastructure. They worry about sudden restrictions disrupting their access to frontier AI models. This incident pushed political leaders to push for European AI sovereignty.

The European Commission is now assessing how export controls like this one could impact global AI collaboration. Some lawmakers want to accelerate funding for homegrown AI models. But building models that match the scale and power of U.S. frontier AI will take years and huge resources.

For companies and AI teams, this ban raises new questions about compliance and risk. Many rely on cloud-hosted AI models across borders. Sudden shutdowns or access blocks can cause major disruptions. Teams must now review contracts, data flows, and prepare for uncertain regulations.

Microsoft CEO Satya Nadella recently said that winning with AI means building internal “learning loops” to improve workflows and judgment, rather than just relying on the best models. This ban illustrates the risks of depending too much on a few centralized AI providers controlled by governments.

In short, the Anthropic ban shines a light on the fragile balance between national security and innovation. It shows how government actions can unintentionally weaken cybersecurity defenses. It also highlights the growing push for transparent AI regulation based on science and fair processes, not politics.

Security experts are calling for smarter policies. Policies that protect the public without hampering defenders or slowing progress. Until then, the fallout from this ban will continue to ripple through AI research, cybersecurity, and global tech cooperation.