JustSayAIWeeklyReport

“Let me just say this first—if you look at these three stories this week on their own, one is Google building a data platform with the UN, one is NATO putting a small model on drones, and one is Korean prompts testing out as low-harm—they sound completely unrelated.”

• NATO puts small AI models into drones, shifting AI from cloud chat to weapons and trenches.
• UN: sharing; NATO: exclusivity; same technology foundation speaks openness and "I can hit you, you can't hit me."
• Once AI enters national systems as a general-purpose technology, it is first carved into ours and yours.

“What's Google after?”

• A platform’s value isn’t openness itself, but giving models an endorsed coordinate system.
• UN brand is half the product; rest is Google owning default narrative: whoever defines it, their model answers best.
• Open sourcing opens interfaces so others follow Google's schema; the real lock-in is structure.

“The materials say it identifies and attacks on its own, but in real systems there’s usually still a human pulling the trigger—at least nominally.”

• Materials say “autonomously identify and attack,” but real systems still usually have a human pulling the trigger—at least nominally.
• Identification, recommendation, and authorization form three layers; each step forward blurs responsibility one layer, especially procurement responsibility.
• The industry playbook: push capabilities forward, pull wording back; no one signs "authority to fire," but they sign "decision support."