The Pentagon just hired Grok

The architecture wars are here. If language models are already “good enough”—then what comes next? We dove deep into world models vs language models, the architectural divide that could reshape everything from robotics to AGI. Yann LeCun left Meta to chase a $5B valuation on this bet. The billion-parameter arms race? Might be fighting the wrong war entirely.

Meanwhile, the agent race is heating up. We broke down Apple’s pivot to Gemini, Anthropic’s new Cowork feature, and Microsoft’s widening AI gap—three stories that tell you exactly where this is headed. Now the frontier labs are formalizing the rules: Thomson Reuters just launched its Trust in AI Alliance with Anthropic, Google, OpenAI, and AWS as founding members. Self-governance is suddenly fashionable.

And then there’s Washington. The Pentagon’s new AI acceleration strategy is dropping guardrails—and picking Grok as the model to integrate into classified networks. Hegseth’s announcement Monday explicitly sidelines AI ethics in favor of speed. The reasoning? “We’re in a race, and the bureaucracy was losing it.” The private sector wants rules. The government wants results. 2026 is going to be interesting.