Anthropic Blocks Fable and Mythos AI Models Amidst China Access Allegations

Anthropic has abruptly halted access to its Fable 5 and Mythos 5 AI models for all users worldwide, a move reportedly triggered by a US government directive. The shutdown comes amidst escalating concerns that a group with ties to the Chinese government may have gained access to the more advanced Mythos 5 model, raising potential national security risks for the United States.

Initial reports from the US magazine Semafor, citing an anonymous source, suggest this suspicion of Chinese government access is a primary driver behind the decision, potentially explaining why both Fable 5 and the more restricted Mythos 5 had to be suspended. While confirmation is pending, such access could pose a significant threat to US national security.

Por que importa

Fable 5, launched by Anthropic just last week, was introduced as a content-restricted version of the new Mythos 5. Early testing confirmed Mythos 5’s superior performance but also highlighted criticisms regarding its security measures. Anthropic had implemented a system where a classifier would route sensitive queries to an older model, Claude Opus 4.8, instead of Fable 5. This was intended to prevent malicious use for cyberattacks or other harmful purposes. However, this mechanism also reportedly misidentified “safe, normal content” as requiring redirection, leading to user frustration.

The situation escalated over the weekend when Amazon CEO Andy Jassy reportedly informed US Treasury Secretary Scott Bessent about a suspected “jailbreak” of the AI models. This report was allegedly based on an analysis from his company’s cybersecurity department detailing the alleged exploit. The information regarding China’s potential access to Mythos 5 emerged subsequently, contributing to the ongoing lack of clarity surrounding the exact circumstances.

Contexto

Key facts

The implications of this development are significant for the AI community and national security. The incident underscores the ongoing challenges in balancing AI model capabilities with robust security and export control measures. For developers and researchers, it highlights the potential geopolitical ramifications of advanced AI technologies and the stringent regulatory oversight they may face. The US government’s directive suggests a proactive approach to mitigating risks associated with AI models that could be exploited by foreign adversaries.

The sudden global shutdown of Fable 5 and Mythos 5, even for foreign employees of Anthropic within the US, points to the broad scope of the export control directive. This action by the US government, if confirmed to be related to national security risks posed by potential access from China, could set a precedent for how AI model access is managed and regulated internationally. Further details are expected as investigations continue.

Source: Heise KI – Sperre von Anthropics KI-Modellen: China soll Zugriff auf Mythos gehabt haben – https://www.heise.de/news/Sperre-von-Anthropics-KI-Modellen-China-soll-Zugriff-auf-Mythos-gehabt-haben-11331767.html