New York police report warns AI powered drones raise urban risks

The drone report captures how quickly off the shelf hardware plus increasingly competent AI is turning into a domestic security problem. Most of the systems described are not AGI level by any stretch: they are pattern recognition, navigation and simple autonomy layered on cheap quadcopters. Yet even that mix already produces thousands of illegal flights over stadiums, credible worries about swarms targeting soft infrastructure and a scramble by local police to learn new counter‑UAS authorities. ([monorepo-sample2.nyt.net](https://monorepo-sample2.nyt.net/2026/09/19/nyregion/artificial-intelligence-drones-police.html))

For the AGI race, this is a preview of the “narrow but dangerous” phase. Long before a general system exists, relatively simple models will give small actors the ability to script and coordinate physical attacks at scale. That, in turn, will push governments toward more aggressive surveillance, more automated defenses and potentially more pressure on labs to build defensive AI tools that can keep up. If the only answer to AI enabled threats is more AI, we could end up with a tightly coupled offense defense spiral around autonomy and targeting. That environment is unlikely to be friendly to careful, measured AGI development.