NASA used Claude AI to plan Mars rover drives in December

By Splitfeed EditorialLast updated Sep 28, 2026, 11:10 UTC

In December 2024, NASA's Jet Propulsion Laboratory used Anthropic's Claude models to help plan two drives for the Perseverance rover on Mars. Human planners reviewed and adjusted the AI-generated routes before any instructions were uploaded to the rover. Separately, in May 2024, NASA and IBM deployed a compressed AI model aboard the International Space Station and a satellite, tasked with identifying surface and atmospheric features such as floods and clouds directly from orbit. Both deployments were reported by IEEE Spectrum.

The significance of these deployments goes beyond novelty. Mars operates under communication delays that can stretch to tens of minutes each way, which has historically made engineers cautious about delegating any decision-making to automated systems where errors could be mission-ending and uncorrectable in real time. Using a large language model like Claude to assist with route planning, even with humans retaining final approval, represents a meaningful step toward integrating commercial AI tools into mission-critical workflows. The onboard IBM model on the ISS and satellite goes further, performing inference directly in orbit rather than relaying raw data to Earth for analysis, which reduces bandwidth demands and response time for time-sensitive observations.

No published reactions from the Pro-AI, Anti-AI or Middle Ground camps have appeared for this specific story. A Pro-AI camp would typically highlight these deployments as proof that AI is ready to assist in the most demanding real-world environments, pointing to the human oversight loop as a responsible design rather than a limitation. An Anti-AI camp would likely focus on the risks of embedding commercially developed models into safety-critical government missions, questioning what validation standards apply and who bears accountability if a route recommendation causes harm. A Middle Ground camp would probably welcome the human-in-the-loop structure for the rover drives while pressing for published benchmarks on the onboard model's accuracy before broader autonomous deployment is considered.

The details most worth watching are whether JPL expands Claude's role beyond the two December test drives to routine mission planning, and whether NASA releases any performance data on the IBM orbital model's feature identification accuracy across varied conditions. Those disclosures would give observers a concrete basis for assessing how far autonomous AI decision-making in space has actually progressed.