Why AI news got a lot messier this week

AI news is getting weird fast: leaked model features, Pentagon deals, and employee pushback all hit at once. Here are the must-know shifts shaping the week. ^[1][11][2]

Pentagon AI deals: Google joined OpenAI, xAI, Nvidia, and Microsoft in allowing models on classified networks for "any lawful purpose," while Anthropic refused. The fight is now about guardrails, not just model power. ^[11][1]

Google backlash: hundreds of employees signed an open letter over the Pentagon deal, but leadership is standing firm. The old Project Maven playbook may not work anymore. ^[11]

AI in medicine: Harvard research highlighted by AIFlashReport says AI offered more accurate ER diagnoses than two human doctors. If that holds up, frontline triage could change fast. ^[1]

Google Veo 3 is now widely available, while The New Yorker says AI firms still face the same hard question: how do they make real money at scale? Product adoption is racing ahead of the business model. ^[5][10]