Asteroid-mining startup AstroForge has looked at the plot of every killer-AI film ever made and thought: yes, but faster. Its 2027 mission, Autonomy-1, will hand full control of a spacecraft to an in-house AI called Solo and, for good measure, carry no radio capable of receiving commands from Earth. Once it separates from the rocket, no human can tell it anything. Not “stop.” Not “turn left.” Not “please don’t.” It is, per the company’s own boast, the first space mission designed to finish itself without a single instruction from the ground.
The reasoning is somehow both sensible and unhinged
Here is the bit that stops it being pure hubris: they have a reason, and it is a good one. In 2025 AstroForge lost its Odin probe, and traced the failure not to the spacecraft but to the link with Earth. So the logic goes: if the phone line is what keeps killing you, rip out the phone. At asteroid distances, waiting minutes for mission control to reply while a problem compounds is its own way to die, so an onboard brain that just handles it removes a real, fatal weakness. Fine. Defensible. Genuinely clever, even.
And then they removed the ability to intervene at all, which is where the engineering elegance tips into the kind of decision that gets a chapter named after it. Solo does sit on top of proper deterministic, physics-based flight software, and it was trained on data from roughly 2,500 spacecraft sensors, and it will fly in “shadow mode” watching before it takes over. So it is not a language model in a trench coat steering a rocket on vibes. It is more considered than that. It is also a spacecraft you cannot phone if it decides to do something interesting, which is a sentence that should make any adult pause.
Why it actually matters
Strip away the swagger and this is one of the cleaner tests of a question the whole industry keeps dodging: do we trust these systems to act alone when there is no undo? A misbehaving AI agent on Earth gets switched off. A spacecraft with no receiver, halfway to a rock, does not. AstroForge is betting a capable autonomous model beats a fragile radio, and if Autonomy-1 works, it genuinely rewrites how deep-space missions run. If it does not, it fails in the most spectacular, unrecoverable way available to engineering, with a full audience and no off switch. Bold is doing a lot of work in that plan. So is reckless. As usual, we will only find out which after the fact. (Sources: TechCrunch, Space.com, AstroForge, September 2026.)
Related: AI agents are already loose on real workflows down here, too.