The U.S. government has expanded a ban on Anthropic’s flagship Fable 5 and Mythos 5 models, moving from a targeted restriction to a total block for foreign nationals. The escalation followed reports of security vulnerabilities flagged by Amazon researchers and concerns regarding investor SK Telecom’s alleged ties to Chinese security risks. Anthropic has suspended access to the models globally while negotiating with the President’s AI Council to resolve 'jailbreak' vulnerabilities.
AI infrastructure ambition has moved beyond GPU counts to raw electrical capacity. Meta’s massive 1.6-gigawatt deal with Crusoe this week underscores a new era where tech giants are contracting for energy at the scale of national grids to fuel the next generation of multimodal reasoning models.
Meta's internal 'Applied AI' unit consumed 73.7 trillion tokens in just 30 days, leading to an internal crackdown on 'tokenmaxxing'—a practice where employees inflate AI usage to boost performance metrics.
Which restricted Anthropic model class was developed specifically for cybersecurity defense under the U.S. government's 'Project Glasswing'?