Too dangerous to release: is Mythos the start of the restricted-AI era? (www.nature.com)

Anthropic has made headlines by announcing its Claude Mythos AI model, deemed too dangerous for public release due to its ability to identify vulnerabilities in all major operating systems and web browsers. This significant move reflects a shift towards restricted access in AI research, with experts predicting that other companies, including OpenAI, may soon adopt similar strategies. Anthropic’s limited release of Mythos to about 50 vetted organizations raises concerns over national security and public safety, signaling a potential new era where advanced AI models may not be widely accessible. The implications of this approach are profound for the AI/ML community. While restricting access could mitigate risks associated with misuse, it also challenges the longstanding belief in the importance of transparency and collaboration in scientific research. Some researchers fear that such restrictions will entrench inequality, granting only privileged access to the most powerful tools. This trend suggests a possible regulatory shift where governments may impose stricter controls on advanced AI technologies, similar to those applied to dual-use goods, thereby complicating the landscape of AI research and development for years to come.