On October 10, 2026, SignSplit raised a $400 million seed round at a $1 billion valuation — from a single funder, W Group — for what it calls "signed data": technology to protect and license data, creative work, and personal likeness in the age of generative AI. A $400 million seed for a company that has barely begun is a statement about how big investors think this problem is.
The context is a year of accelerating moves to control likeness and training data: Matthew McConaughey's anti-AI-misuse trademark filings in January 2026, Taylor Swift's similar actions in April 2026, Denmark's AI likeness law in 2026, and California's SB 1050 requiring disclosure for synthetic performers in ads, signed in 2026. The legal scaffolding for a licensing economy is being built in real time.
This article explains what "signed data" means, how the value chain could work, and where the startup opportunities lie.
What "signed data" actually means
The concept has three layers:
Provenance — a cryptographically verifiable record of where data came from and who owns the rights. Standards like C2PA (backed by Adobe, Microsoft, and camera makers) already embed content credentials in images; the idea extends to text, audio, voice, and likeness.
Consent receipts — machine-readable records that a rights holder agreed to a specific use: training this model, generating in this style, for this compensation, in these territories. Think of it as the digital equivalent of a sync license in music.
Detection and enforcement — technical means to identify AI-generated or AI-trained content in the wild. Google's SynthID watermarks AI-generated media; OpenAI has reportedly researched text watermarking. Watermarking doesn't prevent misuse by itself, but it makes provenance checkable.
SignSplit's bet is that combining these layers creates licensable infrastructure — a registry where data and likeness can be permissioned, priced, and tracked.
Who pays whom: the value chain
A functioning likeness economy needs clear money flows:
- Model builders (the labs) pay for training rights — either directly or through aggregators — because unlicensed training is increasingly a legal and reputational liability.
- Rights holders (creators, celebrities, publishers, and eventually ordinary people whose data has value) get paid per license, ideally with usage-based royalties rather than one-time buyouts.
- Aggregators and registries take a cut for assembling licensed datasets at scale, verifying rights, and distributing royalties — the role record labels and collecting societies play in music.
The music analogy is instructive. Sampling went from wild-west appropriation to a licensed, royalty-bearing system — not because artists stopped being sampled, but because the law caught up and infrastructure (clearinghouses, PROs like ASCAP/BMI) made licensing cheaper than lawsuits. The likeness economy is arguably at the pre-clearinghouse stage.
Startup opportunities
Consent infrastructure. The rails for granting, revoking, and verifying consent at internet scale don't exist yet. This is the deepest white space: identity-bound permission records that survive across platforms.
Royalty tracking and distribution. If licensed training data generates royalties, someone must attribute model outputs back to inputs and cut checks. This is computationally hard (which training examples influenced which output?) but commercially valuable.
Identity and likeness registries. Verified registries of who owns which likeness, voice, and style rights — the "whois" of the likeness economy.
Enterprise compliance tooling. Companies deploying AI need to prove their training data and generated outputs are clean. Expect demand for audits, provenance dashboards, and indemnification products.
The risks: unverifiable claims and jurisdictional fragmentation
Two structural risks deserve honest treatment:
Unverifiability. A "signed" dataset is only as trustworthy as its signer. If anyone can mint provenance claims, the system needs reputation and dispute-resolution layers — otherwise it's DRM with better marketing. Investors should diligence how a startup verifies the rights it claims to license.
Jurisdictional fragmentation. Denmark's likeness law, California's SB 1050, and whatever comes next create a patchwork. A global model trained under one regime's rules may violate another's. Startups building consent infrastructure must design for the strictest plausible regime, not the friendliest current one.
Why it matters for founders and investors
For founders: If your product touches training data, generated likeness, or synthetic media, build consent and provenance in from day one — retrofitting is far more expensive. The compliance tooling category is greenfield: every enterprise deploying generative AI will soon need answers to "where did this come from and do we have the rights?"
For investors: The $400 million seed tells you the smart money believes licensed-data infrastructure is inevitable. But the winners won't be whoever raises the most — they'll be whoever solves verification and cross-jurisdiction enforcement. Diligence the rights-verification mechanism, not the valuation.
Angeled.com connects founders with investors.
Read Next
Open Weights Are Not Open Business Models: Mistral Large 4 and the Licensing Trap
Courts Are Writing AI's Rules Now: Three Rulings Founders Should Read
Delhi High Court Orders 24-Hour Takedown of AI Deepfake Videos Targeting Chief Minister
EU Court Adviser Says AI Training Is Legal — but Gemini's Answers May Still Infringe Copyright
AI Regulation After the Midterms: What Founders Should Do Before the January Bills Arrive
DOJ Launches Super Intelligence Task Force to Police AI Misuse



