[Instrumental Open. No Vocals] source familiar with Anthropic’s gov work told NBC Claude don’t “pull the trigger” it’s “just” decision support, see not an autonomous targeting engine run that back and report, see decision support system not autonomous targeting engine, they say the word “support” doing heavy labor in that phrase Maven eats the satellite feeds all the drone footage streams SIGINT hits, geolocation pings raw data, raw beams Claude reads it synthesizes, ranks, sorts spits out target lists with laser‑tight GPS cords weapons recs auto‑generated strike reports legal briefs pre‑filled to make the killing look in‑court thousand names in the first twenty‑four but Claude does not “recommend” it just “supports” the war Paul Scharre at CNAS “AI lets the military build target packs at machine speed not at human pace,” he says clearly the “human” still decides rubber‑stamps at machine speed that’s the “support” that’s the ghost inside the feed Heidy Khlaaf at AI Now speed is sold as strategic but really it’s a mask for mass‑scale, indiscriminate hit lists when you clock how off these models are world policy hub ran statistics decision support and auto‑target gen at scale same pipeline, same code path ethics flipping heads or tails the school in Minab been a school since sixteen system pushed the coordinates paperwork washed the scene auto‑legal justification signed and sealed in the screen a human “approved” it but the timing was machine that was the support that was the decision Claude does not directly offer targeting “recommendations” the coordinates pre‑baked just waiting for permission the law already drafted for the post‑strike revision the human had the button but the support had already made the decision
