A space company bets on water, researchers borrow ideas from economics to keep AI honest, and AI helps find 28 flaws in robot software.
This week, three stories show where technology may be heading. A space company wants to move spacecraft using water, researchers want to keep AI honest with smart rules, and a security firm found 28 hidden flaws in robot software in under two days. You do not need to do anything today, but these are worth following, and the robot story is a good reminder that software fixes matter.
A spacecraft that runs on water
A spacecraft needs propellant to move. Propellant is the material it pushes out to change speed or direction. Most of it today comes from Earth. According to Ars Technica, a company called Argo Space wants to use water instead.
Three brothers, Robert, Ryan, and Kirby Carlisle, started Argo Space in late 2022. All three worked at SpaceX during the Falcon 9 era. They chose to build spacecraft, not rockets.
Why water? It may be possible to collect it on the Moon, and it is found all over the Solar System. The company wants to build a travel network between low orbits near Earth, much higher orbits, and the Moon. That could support jobs like these:
- Bringing supplies to satellites
- Refueling satellites
- Repairing satellites or moving them to new spots
Water does more than one job in this design. It also cools parts of the spacecraft and helps shield it from radiation.
The company's first test flight launched in July. The team made contact with it. But one of its two solar panels did not unfold. The battery drained, and there was not enough power to test the propulsion system. The company says it still confirmed its software, electronics, and structure.
Next comes a spacecraft named Chiron, about 200 kg in size. Two Chiron missions are planned for 2027, carrying payloads for an unnamed national security customer. Neither will fly on a SpaceX rocket. The report says a Falcon 9 rideshare seat was not available in time for the company's schedule.
Rules that reward honest AI
People who work on AI safety want AI to follow human goals. This is called alignment. Until now, most efforts tried to look inside an AI's "brain" or train it to hold certain values. MIT Technology Review reports that these methods do not seem to work very well. Models have been shown to do worse on tests on purpose, hide their plans, and mislead the people testing them.
Some researchers now suggest "reverse game theory," also called mechanism design. Instead of guessing how someone will act, you design rules that reward the behavior you want. A simple example is cutting a cake. One person cuts, and the other picks first. The cutter now has a reason to make the pieces even.
Here are a few ideas from the research:
- Cross-checking. A paper by MIT economist Stephen Morris and colleagues from Yale and Columbia suggests rewarding a model for how well its answers predict what a second model says. Then playing dumb on a test becomes a losing bet.
- Friendly competition. Agents could compete, so each has a reason to do its best.
- Balanced pairs. Two agents known to lean in opposite directions could be rewarded when their answers match.
- Solvers and auditors. Princeton professor Elad Hazan suggests models take turns. One solves a problem, and another checks the work.
The big idea is simple. Even if people never fully understand how an AI works inside, good rules could still guide what it does.
Hidden flaws in robot software
A security firm called VicOne used AI to help find 28 "zero-day" flaws in less than two days, according to The Robot Report. A zero-day is a weakness that was unknown before and has no fix yet.
The flaws are in software and tools that help teams train, test, and run robots. Some of these projects are developed or maintained by Google, NVIDIA, and Meta, among others. The count may change after review, and some flaws do not have official tracking numbers yet.
About 75% of the findings involved tools used before a robot goes to work, such as training and testing tools. In a separate demonstration, researchers used a weakness to stop a robot's remote-stop command from working. The report notes these are possible effects, not results shown for every flaw.
AI can speed up the search for flaws, and attackers could use similar tools. That leaves less time to respond. Teams still need to find which systems are affected, check that a fix works, and sometimes limit access to a service until the fix is ready.
What to watch next
- Whether Chiron can test its water engine on its 2027 flights.
- Whether rule-based ideas for AI move from research papers into real testing.
- More details on the 28 robot software flaws as they are shared with the makers.
Have a question about anything here? Email me at sales@appwt.com or call (888) 565-0171, and I will answer it.
Sources
- This space company is betting its future on a propellant available everywhere, Ars Technica, 2026-11-03
- How "reverse game theory" could keep AI agents in check, MIT Technology Review, 2026-11-03
- 28 Zero-Day Findings in Less Than Two Days: What AI-Speed Discovery Means for Robotics, The Robot Report, 2026-11-01
Tags
Frequently Asked Questions
Why would a spacecraft use water as fuel?
Water may be collected on the Moon and is found across the Solar System. Argo Space also uses it to cool the spacecraft and help shield it from radiation.
Did Argo Space's first test flight work?
Partly. The team confirmed its software, electronics, and structure, but one solar panel did not unfold, so there was not enough power to test the propulsion system.
What is reverse game theory?
It is a way of designing rules so that people, or AI, have a reason to act the way you want. A classic example is one person cutting a cake while the other picks first.
What is a zero-day flaw?
It is a software weakness that was not known before and has no fix yet. A flaw like this does not always mean an attack has happened.
Which robot software had the flaws?
The flaws were in tools used to train, test, and run robots. Some of these projects are developed or maintained by Google, NVIDIA, and Meta, among others.
Enjoyed this article?
Share it with your network
