W Group puts $400 million behind SignSplit's wager that AI developers will pay for consented data

SignSplit's own account suggests that when legal asks an AI team where the voices in a training set came from, the usual answer is the open web ^[17]. The company says most existing AI models make do with scraped data, much of it of opaque origin. It expects developers will likely have to pay for better human data with provenance, permissions and usage rights attached ^[17].

Scraping is what users do. The company and its investor are counting on regulation to change that. W Group's announcement cites the EU AI Act and a growing set of US federal and state laws and proposals on privacy, biometrics, deepfakes and digital likeness. It argues they move the standard from "was the data available?" to "can you prove the person agreed?" ^[18] Volodymyr Nosov, W Group's founder and president, compared it to payments. "Digital finance scaled once people could trust that every transaction was authorized, recorded and settled," he said. "The AI economy needs the same for data: proof that a person agreed, a record of every use and fair compensation for what they contribute." ^[15]

The pitch runs well ahead of the product. "We believe that every serious participant in AI and the data economy will come to rely on SignSplit's infrastructure," Nosov said ^[16]. The launch materials do not name an AI developer buying from a pool, the number of people who have signed their data, a price, or how much of the $400 million is cash and how much is strategic resources. Glib Denisov, the co-founder, executive chairman and chief product officer ^[13], described the company as built ahead of demand. "When we started SignSplit in 2024, we were building for where we believed AI was going, not where it was at the time," he said ^[12].

W Group's cheque equals 40% of the $1 billion valuation ^[14]. The group calls itself a fintech and technology ecosystem serving more than 40 million users ^[10]. SiliconANGLE describes it as 11 financial technology firms focused on blockchain ^[11]. Nosov said the group will add capital and resources to speed up a global rollout ^[9].

For the data lead deciding whether signed data becomes a line item, two tests sort it. The first is whether the data can be scraped at all. The second is whether it involves faces, voices or likeness, the categories covered by the laws W Group cites ^[18].

Data that cannot be scraped and does involve a person's face or voice, such as a voice model trained on consenting speakers or a study that needs real participants, is where I would budget for signed data now. SignSplit's research pools for universities and life sciences firms are built for that quadrant ^[6]. Data that cannot be scraped and involves no one's likeness, such as robot task demonstrations, is a collection purchase, and the consent record comes with it. Scraped faces and voices are the quadrant the regulatory pitch targets. Paying there is a bet on how quickly the cited rules are enforced. If a team waits and the rules arrive early, it has to relicense on the regulator's timetable. Scraped text with no likeness stays scraped. That team has no reason to pay yet.