Of course Sigint want to be letting these models scan everything but I guess the really interesting part is how much data the NSA has to make the models even better.
Not sure how valuable that data would be to train models. I’m guessing it’s millions of routine phone conversations and text messages. Maybe a ton of satellite imagery. I wouldn’t consider that to be particularly useful to go from gpt5 to gpt6.
I was thinking industrial espionage and secrets/technology not publicly available but I’d be surprised if the made a database of it and uploaded it to the cloud having though about it more clearly.
Very typical move of a company planning to enter the military supplier space.