Full Summary
This Saturday morning, Anthropic has globally suspended its two most powerful AI models, Claude Fable 5 and Claude Mythos 5. Both Reuters and the BBC confirm this action follows a US government directive citing national security concerns. The government order initially aimed to restrict access for foreign nationals, but Anthropic stated it was impossible to filter users by nationality in real-time, leading to a full worldwide shutdown. NewsBytes and ThePrint highlight that this marks the first time a leading AI company has taken a publicly deployed model offline due to federal intervention. The core concern, according to multiple sources including Traders Union and ThePrint, is a potential "jailbreak" of Fable 5 that could bypass its safety controls. Anthropic, however, argues this "jailbreak" involves prompting the model to identify software flaws, a capability already available in other public models and routinely used for defensive cybersecurity. The company also noted the government provided no specific details on the national security threat. In other AI news, India has launched Varya AI, a new video generation model from Avataar. Techlusive reports it creates videos from simple prompts, tailored for Indian users by understanding regional cultures and festivals. Meanwhile, Arabic.AI and Stanford University released the HELM Arabic Enterprise benchmark to standardize evaluation of Arabic AI systems. Google's Gemini 3.5 Pro is also set to launch later this month, boasting a two-million token context window and a "Deep Think" feature for complex reasoning. The Anthropic shutdown sets a new precedent: government permission may now be needed to release frontier AI models. This could significantly impact how advanced AI is developed and deployed, potentially affecting the pace of innovation and the availability of powerful tools for businesses and individuals alike.