A SpaceX Falcon rocket lifted of from Space Launch Complex 40 (SLC-40) to begin the NROL-69 mission on behalf of the National Reconnaissance Office and the U.S. Space Force’s Space Systems Command. Image: Michael Cain/Spaceflight Now
By Sandra Erwin,
Published by Space News, 2 November 2026
WASHINGTON — A senior U.S. intelligence official has raised the possibility of designing future spy satellites around the needs of artificial intelligence rather than human imagery analysts.
Didi Kuo, principal deputy director for geospatial intelligence at the National Reconnaissance Office, said Oct. 29 that advances in AI could challenge longstanding assumptions about the image quality required from intelligence satellites.
Speaking at Silicon Valley Space Week in California, Kuo suggested that satellite designers could start by determining what information AI algorithms need to identify objects and detect activity, then build spacecraft capable of collecting that information.
“What would happen if we … built the algorithms first, understood what they needed, and then built the satellite based on what they need versus a human?” Kuo said.
The NRO, which builds and operates U.S. intelligence satellites, has demanding image-quality requirements intended to support human analysts. Those requirements can influence the size, complexity and cost of spacecraft.
“Traditionally, we’ve built high quality satellites for humans. We have something called a NIIRS scale, which is based directly on a bunch of humans, and then we build our satellites on that, and that can drive costly satellites because humans want the best,” Kuo said.
The National Imagery Interpretability Rating Scale, or NIIRS, is an intelligence community standard that measures image quality according to the level of detail a human analyst can distinguish.
Kuo questioned whether satellites intended primarily to feed automated intelligence systems would need to meet the same standards.
He said engineers could determine the minimum sensor performance required for an AI model to carry out a specific intelligence task.
That approach could make it possible to use less expensive sensors and spacecraft while deploying more satellites to provide frequent coverage of areas of interest.
Kuo compared the sophistication of some satellite designs to the James Webb Space Telescope, NASA’s multibillion-dollar infrared observatory, arguing that spacecraft optimized for machine analysis might not require similarly elaborate technology.
“My hypothesis is, as we get more advanced, that those satellite costs and designs don’t have to be as fancy as a JWST,” he said, referring to the telescope.
The approach could also involve trade-offs. Satellites optimized for particular automated detection tasks might not produce imagery suitable for the full range of analysis that human intelligence specialists perform.
“It’s about getting a minimum viable product … but then the question is, who’s that for? A human being or a machine?” he said.
“I’m not sure I’m very popular at NGA when I design a satellite for a machine, but certainly I think that’s an opportunity we have,” Kuo said, referring to the National Geospatial-Intelligence Agency, which specializes in the analysis imagery and other geospatial intelligence for military and intelligence customers.
See: Original Article
