Why Cyber and Bio-Terror Concerns Sparked the Claude 5 Ban-And Why It Was Just Lifted

The government has lifted its restrictions on Fable 5 and Mythos 5, two of Anthropic’s most talked-about and most powerful models. The company launched both models on June 9, 2026. Within just 72 hours, the government slapped restrictions on them.
Public access to the two models was cut off through the U.S. Department of Commerce and a national-security executive order from the Trump administration, on the theory that they could be used for cyberattacks, dangerous biological research, and the development of advanced military technology.
China and Russia could easily put these models to military use. On top of that, by opening up millions of fake accounts, Chinese AI systems could just as easily use Fable 5 and Mythos 5 to train their own models! That risk is very real.
U.S. national security agencies have legitimate, complex, and technical reasons to be concerned about Claude Fable 5 and Mythos 5. It was these concerns that prompted the temporary ban in the first place.
Unlike other models, these don’t just answer a prompt and stop there. Instead, they can work through one subtask after another and carry an entire complex project through to completion. They can even verify much of their own information.
The U.S. government’s fear was this: if someone hands the model a harmful instruction — say, “examine every security aspect of a particular website, probe it for weaknesses, and hack it” — it would keep trying, attempt after attempt, entirely on its own, without any human help. This is what’s being called an autonomous cyber threat.
A seamless fusion of malware and cyber weapons might become possible. Because these models are so powerful at coding, they could be used to build malware that no conventional antivirus system may even detect — malware capable of paralysing highly sensitive infrastructure such as the U.S. defence system, the power grid, the banking system, and water and energy supply networks.
There are concerns about the models’ Chemical, Biological, Radiological, and Nuclear capabilities. The scientific and reasoning abilities of these fifth-generation Claude models (as evidenced by a GPQA Diamond score of 92.6%) rival those of graduate-level scientists. In the wrong hands, they could enable even ordinary labs to create complex synthetic viruses, chemical formulas, and biological weapons.
Although Fable 5 and Mythos 5 are built the same way, when it comes to use and safety, Mythos 5 is the more worrying of the two. That’s why it has been cleared for use by a handful of defence firms approved by the U.S. government.
According to reports, Fable 5 features an automatic safety override. Whenever it processes restricted or unlawful commands, the system downgrades operations to the lower-tier Opus 4.8 mode.

The U.S. government wants to hold on to its top spot in AI technology, while China wants to push the United States aside and strengthen its own position. So far, the U.S. leads on every front in AI. The worrying part, though, is that the gap is narrowing. Meanwhile, Russia, Japan, Singapore, and even the United Arab Emirates are pouring huge sums into AI. And the European Union has no intention of falling behind in the race either.
Anthropic is now working more closely with the U.S. government to bring these security risks down — even though it’s nearly impossible for any AI to be 100% safe.
For all these worries and anxieties, the reassuring news is that people will be able to use Fable 5. And we should hold back from sharing our sensitive personal information with powerful AI models like these — things like bank account numbers, credit card numbers or photos of our cards, and passwords for any email or account.
Source: Reuters, BBC