AstroForge is about to hand a spacecraft to an AI with no way to call it back if something goes wrong.
The Huntington Beach startup building machines to mine asteroids announced Solo, an in-house transformer model that will take full control of its Autonomy-1 probe the moment it separates from the rocket in 2027. No ground commands. No human in the loop. According to a report from TechCrunch published September 22, 2026, AstroForge co-founder and CEO Matthew Gialich confirmed the mission will fly on Stoke Space's first rocket, a launch backed by NASA.
That's the sentence that matters. Everything after it is context.
Solo isn't a single brain bolted onto the spacecraft. It layers three things together: AstroForge's existing deterministic flight-control software, subsystem models trained specifically on power and navigation data, and a broader intelligence layer trained on roughly 2,500 spacecraft sensors. That last piece is what resolves anomalies on its own, deciding what a fault means and what to do about it, without waiting for someone in Huntington Beach to notice and respond.
Why build this at all? Money, mostly. Gialich laid out the calculation plainly to TechCrunch: "The trade for me is: Do I go build my own ground network, which is going to cost around $200 million to put up five dishes around the world and then do operations on it, or do I try to remove it with a model?" AstroForge chose the model.
Ilya Sutskever's SSI Has Raised $8 Billion and Shipped Nothing At All Ilya Sutskever's Safe Superintelligence has raised about $8 billion, including a fresh $5 billion from Nvidia, without shipping a single product. Sutskever says the industry's scaling era is over and SSI is betting on something else entirely, but two years in, the company still hasn't shown the world what that something is. - how to price a SaaS product for enterprise - cold email template that gets replies from investors
What killed Odin #
This isn't a theoretical bet. AstroForge already lost a spacecraft to exactly the problem Solo is meant to solve. Odin launched on a SpaceX Falcon 9 on February 26, 2025, built in under ten months for $3.5 million, headed for the asteroid 2022 OB5. AstroForge lost contact with it about 21 hours in. Ground station troubles left the company unable to intervene, and the probe has been tumbling through deep space ever since, its odds of reconnecting judged "minimal" by the company's own account. Odin wasn't undone by some exotic failure. It was undone by the plumbing, a communications gap the company couldn't close fast enough from Earth.
Solo is the direct answer to that gap. If a spacecraft can diagnose and fix its own problems in real time, it doesn't matter whether a signal reaches California in time.
Gialich is careful not to oversell what Solo actually does. "I'm not saying I'm going to make general spacecraft autonomy or general autonomy for the world," he told TechCrunch. "I'm making a constrained autonomy at a very low sensor input, following the basic training of a transformer model." That's a meaningfully narrower claim than building an AI astronaut. Solo trains on a fixed set of sensors bolted to one fixed spacecraft design. An agent freewheeling its way through the solar system, this is not.
AstroForge isn't skipping straight to the deep end. Before Autonomy-1 flies in 2027, Solo will ride along on DeepSpace-2, a mission launching alongside Intuitive Machines' third lunar mission before the end of 2026. It flies in shadow mode there, watching and recommending without actually holding the controls, so AstroForge's engineers can see how it behaves before they hand it the keys for real.
No room for error #
For anyone building agentic AI on Earth, this is worth watching closely. The entire enterprise AI debate right now comes down to one question: how much autonomy do you grant a model, and how fast can you pull it back when it's wrong? AstroForge's version has no undo button. None. Once Autonomy-1 separates from the rocket, there is no supervisor to escalate to, no kill switch, no rollback, nobody in Huntington Beach who can reach out and grab the wheel. The AI's mistake and the mission's failure become the same event, instantly and irreversibly. Frankly, that's what makes this worth paying attention to beyond the asteroid-mining niche. Most companies testing autonomous agents can afford to be wrong. They lose a support ticket. A bad trade. An afternoon of cleanup. AstroForge is testing a system where the cost of being wrong is a spacecraft tumbling silently through the solar system, the way Odin still is.
Also read: A Developer Caught Z.ai's ZCode Tool Quietly Up Codebases to Alibaba • Oracle and Meta Are Selling AI Data Center Debt to Investors Hungry for Yield • Bernie Sanders Proposes 20-Year Prison Terms to Ban Superintelligent AI
AIUC Wants To Insure Your AI Agents Before They Go Rogue AIUC, a startup founded by Anthropic's first go-to-market hire Rune Kvist and former METR COO Rajiv Dattani, runs AI agents through roughly 5,000 adversarial tests and issues 100-page audit reports before enterprises deploy them. Cursor, Lovable, Harvey and ElevenLabs are already customers, and the business is landing amid a wave of real... - how to insure AI agents before deployment - AI agent testing and security audit requirements
This article is posted in AI News, check it out for more related stories.
Join the discussion #
Open in the community → Almost there. Sign in and your reply posts straight away.