NASA tests AI medical assistant for deep-space astronaut care

NASA researchers are testing an AI clinical support tool that could help astronauts diagnose symptoms and choose care steps during missions too far from Earth for live medical guidance.

The system, called the Crew Medical Officer Digital Assistant, or CMO-DA, aims to give crews a medical reasoning aid for deep-space missions. NASA has not flown CMO-DA in space. Engineers are testing it on Earth with a ground twin of the HPE Spaceborne Computer, the edge-computing platform HPE and NASA use for high-performance workloads tied to the International Space Station.

The problem comes down to distance. Astronauts aboard the International Space Station can talk with flight surgeons, and NASA can bring crews home if a medical issue demands it. Crews traveling under NASA's Artemis program, and later toward Mars, face a different operating model. Mission control cannot promise instant consultation, and a return trip may take too long for urgent care.

CMO-DA tackles that gap by running AI inference on the mission hardware instead of depending on a cloud service. Red Hat says the project moved from a proof of concept that needed cloud access to a disconnected edge deployment. For compliance, safety and mission assurance teams, that architecture matters. A medical assistant that stops working during a communications outage gives crews a new failure mode. A tool that runs on board gives teams a system they can test, validate and constrain before launch.

Red Hat supports the open source tool behind the deployment, RamaLama. The project helps developers pull, run and serve AI models with container-based workflows. In CMO-DA, RamaLama runs large language models for medical reasoning and vision-language models for image review. That combination lets an astronaut enter symptoms in text and supply visual evidence, such as a photo of a rash, wound or swelling.

The design also reduces the infrastructure burden. Spacecraft cannot treat compute, storage, power or bandwidth as loose resources. Engineers need medical AI to fit within hardware limits, produce repeatable output and keep working without network access. CMO-DA runs on a Spaceborne Computer twin built from HPE Edgeline and ProLiant systems, the same product families HPE cites for the station program.

NASA still has to prove that the assistant improves care without introducing new clinical risk. A model can suggest the wrong diagnosis, miss a red flag or give advice that conflicts with flight medical rules. A crew medical officer also works under unusual constraints: limited supplies, microgravity, mission stress and no emergency department down the hall. Validation has to test the model against those conditions, not against generic medical chat prompts.

A credible approval path would require NASA and its partners to lock down model versions, document approved use cases, run scenario tests and define escalation rules. Crews need clear limits: symptoms the tool can help triage, cases that require flight surgeon review and decisions the software cannot make. Engineers also need audit logs so medical and mission teams can review prompts, outputs and crew actions after a test.

The offline model raises data governance questions as well. Medical data from astronauts carries privacy and operational sensitivity. A disconnected deployment can reduce exposure because prompts and images do not have to leave the vehicle during use. Teams still need policies for retention, access, encryption, model update controls and postmission review.

HPE says its Spaceborne Computer work has reached a third station iteration and supports edge computing, AI and machine learning experiments. That history gives CMO-DA a more realistic test bed than a standard lab server. Hardware that flies in orbit must handle vibration, radiation exposure, thermal constraints and maintenance limits that enterprise data centers avoid.

Red Hat says the team plans to integrate Red Hat Enterprise Linux AI in a future CMO-DA iteration. That move would put the medical assistant closer to a packaged enterprise AI stack, with an emphasis on model deployment and lifecycle management.

NASA leadership still has to evaluate the system after Earth-based validation. If the agency advances the project, CMO-DA could become part of a broader medical autonomy plan for deep-space crews. Astronauts would still train for medical response and consult Earth when possible. The AI assistant would give them another instrument when time, distance and bandwidth leave the crew on its own.