A flight recorder for wayward AI systems. The space industry’s record investments. AI spending warnings. All that and more in this week’s Prototype. To get it in your inbox, sign up here .

As an experiment, I built an AI agent trained on my taste that, every Friday, goes through the week’s new album releases and reviews to recommend new music to me. It doesn’t stop me from seeking out human reviews on my own, but I’ve found it to be a nice jumping-off point. (It recommends long reads about music to me, too.) Today, I was excited to see albums that were not on my radar at all from some of my favorite bands. New music from The Molotovs, The Lemon Twigs and the Sheepdogs all in one week? It sounded too good to be true… because it was. My entire briefing this week was completely made up because it experienced an error trying to search the web, couldn’t find new albums, and so made a bunch up based on its training data.

All that is a long way of saying: An AI system can be subject to drift, error and attack, but will often try to keep doing what it’s supposed to do anyway. In music recommendation, an error really only brings disappointment. But for autonomous drones, cars and factories, that kind of failure can be fatal.

That’s where Ashley Reeves comes in. He’s the CEO and founder of South Korea-based ArbaLabs, which is developing AI verification hardware that can record what embedded physical AI systems are actually doing–versus what they think they are doing. “We look at it as a kind of ‘black box’ for physical AI,” he told me, referring to the flight recorders found on planes that can be used to determine what happened in an accident.

The way it works is that its solution, ArbaEdge, is installed directly beside an AI processor, and the chip generates a “digital mirror” of the AI processes being executed. Additionally, the chip protects sensitive data–like telemetry–by outputting encrypted data that verifies results without giving away sensitive information, Reeves said. This can be used by operators to verify that its AI systems are working as intended, and provide a notification if it’s not. Interestingly, ArbaEdge doesn’t use AI itself, Reeves told me. “Everything is rules-based and very auditable,” he said.

ArbaLabs was founded in May 2025, and less than two years later, it already has a proof-of-concept. Literally. Its first demonstration system is currently installed on a satellite owned by partner Space & Bean, launched by a Korean Nuri rocket into orbit on October 7. They’ve established two-way ground communication with the ArbaEdge module on board, and will continue to test its capabilities over the next few months.

In the meantime, Reeves says, the company is currently seeking capital to support continuing ArbaEdge’s development. It’s working with the British Standards Institution to establish global standards for AI verification, and is working on designing its own chip for its next generation of hardware, which it could then license to potential customers. Reeves sees space as a “beachhead” for his company, but plans on moving into terrestrial sectors such as drones, manufacturing plants, autonomous vehicles and medical devices.

“The goal of this company is to be a standardized de facto layer for the entire industry,” he said.

Got questions? Comments? Story tips? Human-generated music recommendations? Email me at: theprototype@forbes.com

Space’s $22 Billion Quarter

It’s been a big year so far for the space industry, as venture firm Space Capital reports that investors poured more than $22 billion into space startups in the third quarter of 2026–which puts the year at $92.5 billion invested, already ahead of 2025.

One thing to note: As I’ve discussed here before , the biggest bottleneck facing the space industry isn’t application or demand, it’s launch. But investors have taken notice and realize that the industry can’t rely on SpaceX to fix the problem. Just $7.8 billion has gone into pure launch companies this year, with Stoke Space topping this quarter’s list with a $1 billion series E round valuing the company, which is building reusable rockets, at $10 billion.

Another watchword: space sovereignty, which the report describes as “the strongest tailwind in the space economy.” The Trump administration’s isolationist instincts have spurred European and Asian nations to push more into their domestic space capabilities. European startups accounted for $6.3 billion of total Q3 investment, reaching $13.5 billion year to date. Much of that spending is focused on defense and launch capabilities.

Now if you were wondering to yourself, “space is great and all, but what about AI?” Don’t worry–AI is getting investor interest in space just like it is on Earth. There’s money going into companies producing geointelligence data, such as satellite images, that autonomous systems use to navigate the world. But there’s also increasing interest on the industrial side, such as a $1.37 billion round for Hadrian, which is building automated manufacturing systems for aerospace.

The Hot Take: AI Can’t Solve The Antibiotic Crisis

I regularly ask investors for their take on tech trends within their industries. This week, I talked to Henry Skinner , the CEO of AMR Action Fund . The venture capital firm, which has around $1 billion in assets under management, is dedicated to investing in companies developing new antibiotics to combat growing antibiotic resistant diseases that kill around 5 million people annually.

What tech is being overhyped right now?

The idea that AI will solve the antibiotic crisis is overhyped . AI is accelerating discovery, but discovery is only a small part of the challenge. The real hurdle is advancing promising antibiotic candidates through clinical testing, where costs rise sharply, risks compound, and private investment has disappeared. In antibiotics, the expertise, infrastructure, and capital needed to carry drugs from Phase 1 through approval have been eroding for years because the market fails to adequately reward innovation. Investors have left the field, and human talent has followed. Unless policymakers address the underlying market failure, AI will generate promising antibiotic candidates that nobody is willing to finance or able to develop.

What tech or trend should more people be talking about today?

We’re not talking enough about the emergence of continuous pathogen surveillance systems . Wastewater monitoring proved its value during COVID but is quickly evolving in some dynamic ways. One team at Baylor University is using next-generation sequencing to rapidly identify and quantify hundreds of viruses from a single wastewater sample and even flag previously unknown threats. Others are developing systems to monitor the air of a building’s HVAC system to detect pathogens and allergens in real time. We’re moving from a world where we wait for sick patients to show up in clinics or at a doctor’s office to one in which we can continuously watch for emerging threats at a local level and, hopefully, mitigate their impact. That shift could fundamentally change how we target public health interventions and deliver care.

What are we all going to be talking about five years from now?

In five years, everyone will be talking about how best to export biotech innovation from China . The current debate over whether China can compete with the U.S. in drug development is becoming obsolete. We’re watching China transition from a fast-follower producing second-generation products to a global hub of biopharma innovation, producing more medicines more rapidly at lower cost while attracting more investment every year. Policymakers are rightly beginning to grapple with our dependence on China for essential medicines and the vulnerabilities that creates in our pharmaceutical supply chain. But they’ll soon also need to debate how to get the innovations China produces to patients around the globe and how regulators, companies, and health systems adapt to a world where novel therapies increasingly originate in China.

Anthropic vs. The Pentagon, Week 33: Anthropic has filed for a re-hearing of the appellate court decision that allowed the Department of Defense to label it a supply chain risk, which as we discussed last week has led the military to stop using its AI tools. It may be seeking out influence in other ways, too: So far in 2026, the company has hired former Trump advisor Sihao Huang to head its frontier compute strategy and appointed former White House deputy chief of staff Chris Liddell to its board of directors, CNBC reports .

Starlink vs. Ukraine: Earlier this week, Ukraine officials complained that Elon Musk’s SpaceX restricted its use of Starlink for military services without warning. The nation relies on the internet provider for its drone warfare operations. Elon Musk responded that the military and U.S. control those services, not SpaceX, and it was reported yesterday that the restriction order came from the White House. In past weeks, the Trump administration has decried Ukraine’s strikes against Russian energy industry targets, a critical piece of Ukraine’s strategy in the fourth year of its defensive war against Russia’s unprovoked 2022 invasion.

The Ongoing Saga Of AI’s Strained Economics: Tech stocks are falling due to investor concerns over inflation and AI spending–and whether losses can get recouped, especially since Anthropic reported slowing revenue growth. My colleague Giacomo Tognini notes that data center developers are increasingly financing new construction with junk bonds . The Wall Street Journal reports that there’s an enormous gap between potential legal liabilities faced by AI companies and the available insurance to cover it. And finally, Gartner reported that in order to make up for AI infrastructure spending, data center owners “generate a cumulative $8.5 trillion in revenue” through 2030. Should be easy enough, right?

AI Side Effects: The enormous AI buildout is having its own ripple effects: Global smartphone shipments declined 20% year over year this quarter, in part because of AI-driven shortages in memory chips. Economist Paul Krugman notes that spending on AI construction is crowding out everything else . And BlackRock CEO Larry Fink warns that pausing data center construction could spark a recession.

Good AI Stuff: Not all is doom and gloom in AI. Colossal Biosciences unveiled an AI model trained on modern bird species it’s using to try and bring back the Dodo. And researchers at MIT developed a machine learning technique to double the accuracy of six-week weather forecasts.

Quantum Progress: Google published results in Science showing that it has found a practical way to produce “magic states” in a quantum computer, which is a crucial step to practical, universal quantum computing. The company has a nice writeup of their solution here . Meanwhile, London-based Orca Computing launched the PT-3 quantum computer, which is designed for “hybrid” computing functions alongside a classical computer. And Germany-based Saxon Q showed its room-temperature quantum processor performs on par with a superconducting quantum computer, noteworthy because one of the latter’s big challenges are the bulky, cryogenic solutions they need to operate at near absolute zero.

What’s Entertaining Me This Week

Today I’m listening to the (very real) album Cookie Happened, the latest from L.A. rockers The Eels. It’s an interesting mix of Americana and rock, moodier than some of the other tracks of theirs that have stood out to me in the past. (“Now You’re Really Living” is one of my all-time favorite songs, and nothing on this album sounds like it.) But that’s okay. It’s a quieter album with songs that draw you in and ask you to pay attention, and I’m happy to give it. Best tracks: “One Razzle, One Dazzle”, “I’m On Standby”, “Dum Dums.”