The US government ordered the shutdown of Anthropic's AI models Fable 5 and Mythos 5, partly because of suspicions that a group with ties to the Chinese government had access to the latter. This is reported by the US magazine Semafor, citing an anonymous source. This could explain why not only the restricted Fable 5, but also Mythos 5 had to be blocked. Access to the latter was actually strictly regulated. If the People's Republic had access to it, it could pose a risk to US national security, Semafor writes. However, confirmation is pending. So far, it was stated that there was a way to bypass Fable 5's security mechanisms, which is why the block was ordered.
Bumpy start and U-turn
Anthropic released Fable 5 last Tuesday as a content-restricted version of the new Mythos 5. Initial tests had confirmed the noticeably higher performance of the AI model but also led to criticism of the security measures. If a so-called classifier deemed a request sensitive, Fable did not answer the question, but the older model, Claude Opus 4.8. Anthropic wanted to prevent attackers from exploiting the model's capabilities for cyberattacks or other malicious purposes. However, this could also happen with “safe, normal content,” according to a note. The approach caused a lot of criticism.
