The Download: dead startups fueling AI and interspecies communication
AI-assisted digest, checked against the original source before publishing. How we work
AI labs are buying the digital remains of shut-down startups for training data, while MIT Technology Review's new Impossible issue scores progress on interspecies communication, lunar cities, and more.
AI training labs are turning to an unusual supply line for data: the digital archives of failed startups. When Keelan Miskell began winding down his crypto startup, he assumed its data held little value. He was wrong—the company's codebase and workspace records have since sold for tens of thousands of dollars to AI buyers.
Miskell is part of a growing market of intermediaries that help shut-down startups liquidate their data for training AI models. Demand has climbed as labs exhaust the supply of public and synthetic datasets, pushing buyers toward real-world operational records of the kind needed to train AI agents. The trade is not without friction: the sales raise unresolved legal and ethical questions about who owns a defunct company's internal data and how it can be used.
The daily must-reads span election infrastructure, satellite internet, and climate. Hackers knocked the US Federal Election Commission's filing system offline a week before the midterms, in the costliest midterm cycle ever, while platforms struggle with AI-generated election misinformation and voting machine software flaws. Starlink restrictions in Ukraine have prompted European officials to rethink reliance on the service, creating an opening for rival providers; an Israeli spy firm has allegedly targeted Starlink users. Greenhouse gas concentrations of carbon dioxide, methane, and nitrous oxide continue hitting records worldwide, even as the IPCC faces institutional pressure. Research on monkeys' brains is also yielding new clues about consciousness.
What to watch: whether the market for defunct startups' data develops clear ownership and consent norms as volumes grow, and whether the animal-communication work crosses from signal cataloguing to genuine interpretation. On the policy side, the durability of election systems and satellite-internet dependence are both moving from technical debates to questions of national and alliance strategy.
Key facts
- • Ex-crypto founder Keelan Miskell sold shutdown startup's codebase and workspace data for tens of thousands of dollars.
- • AI labs increasingly buy defunct startups' data as public and synthetic datasets run dry.
- • MIT Technology Review's Impossible issue scores six long-promised technologies, including interspecies communication.
- • NASA aims to establish a small lunar city by century's end; US and China both pursuing bases.
- • Hackers took the US Federal Election Commission's filing system offline a week before the midterms.
Why it matters
The secondary market for defunct startups' archives gives builders a new revenue path at shutdown and gives labs a scarce asset: authentic operational traces of real work, not synthetic or scraped web text. Expect legal and ethical scrutiny to follow quickly, since workspace and codebase sales implicate employee, customer, and partner data that founders may not fully control. For teams building ag
Spotted an error? Send a correction.
Get the day's AI news in one email
The Daily Epoch · 60-second read · unsubscribe anytime
