AI labs are buying training data from defunct startups

By Splitfeed EditorialLast updated Oct 13, 2026, 09:30 UTC

On October 13, 2026, MIT Technology Review reported that AI laboratories are acquiring data assets from startups that are closing down or have already failed. The central example in the report is Keelan Miskell, who wound down his crypto startup in 2026 and sold its digital archives to AI labs, netting roughly $25,000 for his code base, with an additional payment still pending. Miskell had initially assumed the data was worth little, partly because the startup's users were anonymous and the company had only been operating for a few years.

The practice raises a cluster of questions that sit at the intersection of data privacy, consent, and the economics of AI development. When a startup shuts down, the legal and ethical obligations attached to user data do not necessarily disappear, yet the mechanisms for enforcing those obligations become considerably murkier. Users who interacted with Miskell's platform almost certainly did not anticipate that their activity, even if anonymous, might eventually feed into a commercial AI model. Whether anonymity provides sufficient protection under existing frameworks, or whether something closer to explicit consent is required, remains genuinely unsettled in both law and public debate.

None of the three camps had published reactions to this specific story at the time of writing, so what follows reflects the positions each would typically take. The pro-AI camp would likely argue that purchasing archival data from willing sellers is a legitimate and efficient way to expand training datasets, and that anonymous data carries minimal privacy risk. The anti-AI camp would almost certainly counter that consent was never obtained for this secondary use, that anonymity is not a reliable safeguard, and that the practice creates a perverse incentive to treat user data as a liquidatable asset. The middle-ground camp would probably call for clearer regulatory guidance on data transfer obligations during startup wind-downs, framing the issue as a governance gap rather than an inherently good or bad outcome.

The argument is likely to sharpen as more defunct startups surface as data sources and as regulators in the United States and Europe consider whether existing privacy rules, written before this kind of secondary market existed, apply cleanly to these transactions. Any formal regulatory guidance or enforcement action addressing data transfers during corporate dissolution would go a long way toward settling the core dispute.