What the unit's building, who's joining, and what's next — plus the AI developments shaping the Air Force around us.
The AIU's founding cadre has officially been formed. The unit's first team meeting is scheduled for 12 September 2026 — the moment this goes from a charter on paper to a working team at the table. We couldn't be more excited to get started.
A closely watched gauge of AI model pricing, the LLM Token Expenditure Index, fell to 97 cents per million tokens this week — its lowest reading ever and less than half its peak from earlier this summer. The slide is being driven by cheaper open-source Chinese models like Moonshot's Kimi K3 and OpenAI's own price cuts, and could squeeze margins for frontier labs like OpenAI and Anthropic even as it makes AI cheaper to run for everyone else.
A Russian startup called Mostik has built a way for AI models to exchange the mathematical values inside their weights directly, instead of communicating through ordinary text. The approach has already linked a 753-billion-parameter model with one just 4 billion parameters in size — letting a much smaller model work alongside a giant one without either having to "speak" the other's language.
New York City is imposing a one-year blanket ban on generative AI for students pre-K through 8th grade, citing concerns over child development and AI's effects on kids' mental health. High schoolers will get limited access through five pilot programs, and all students will receive new twice-yearly AI literacy modules covering bias, risk, and how the technology actually works.
A USA Today opinion piece argues against growing calls for a pause on AI data center construction, warning that a US moratorium wouldn't slow global AI development — it would just hand China a structural advantage. The piece frames broad, affordable access to AI infrastructure as the real deciding factor in the race, not raw model performance alone.
The Defense Department has launched military versions of OpenAI's ChatGPT and Grok for employee use, joining Google's Gemini, which rolled out on the Pentagon's AI hub last year. ChatGPT is cleared for unclassified administrative and policy work, Grok is built for operational needs like market research and supply chain management, and more than 1.7 million users have signed up across the three tools so far.