https://spacenews.com/on-the-nros-wish-list-ai-technologies-to-manage-satellites-and-data/

Move fast and break things – was effectively the message the director of the NRO given at an Intelligence and National Security Alliance event. Due to the often time criticality of the NRO’s mission, they are looking to increasingly use artificial intelligence and machine learning for analyzing data on-board their space vehicles, and for orchestrating their overall operations. While recent advances in AI offer much promise for automating data analysis, and solving sophisticated resource allocation problems, it is not the silver bullet most expect it to be.

Adversarial AI in particular has risen in recent years, with researchers and developers showcasing all the ways in which AI systems are brittle and can break. In one instance, researchers demonstrated that by applying a small sticker to a stop sign, a model developed for processing autonomous vehicle imagery data would no longer recognize the stop sign. In other instances, researchers have executed poisoning attacks against AI datasets, corrupting the source used to develop models.

As the NRO seeks to onboard more AI capabilities, much care will need to be given to securing the entire AI development pipeline, from the initial assumptions and data sources used to develop models, to the end models confidence and competence in the field.