MullionAI
Licensed data for AI
The structure behind better models.
Mullion licenses non-public media, pharma and health data to AI labs. Rights-cleared, de-identified and delivered to spec, so owners earn from what they already have.
Three sources, one standard
The data labs can't scrape.
The public web has been used up. The next gains come from data that was never online: studio libraries, decades of pharma research and the clinical records of real care.
Media & content
Your archive has a second buyer.
Film and TV, documentary, sports, music, audio and long-form text, licensed for training on terms you set.
- You choose what is licensed, to whom, and for how long
- Assessed in place, with nothing copied until you approve
- Revenue from content that is already paid for
Pharma R&D
Decades of research, sitting in storage.
Clinical trials, assays, lab notebooks and regulatory reports, turned into compliant datasets for scientific AI.
- De-identified inside your environment before anything moves
- You approve every buyer and every use
- Value from failed and finished programmes
Hospitals & health
Better care data, for better medical AI.
Health records, imaging, pathology and monitoring data, de-identified inside your hospital and licensed under strict governance.
- Nothing identifiable leaves your systems
- Your ethics and privacy teams sign off on every use
- New funding for research and care
Linked end to end
The whole story, not fragments.
A single file teaches a model little. Where the source allows it, we keep each record connected to everything around it, so labs get the full context.
Pharma
- Compound
- Assay results
- Trial outcome
- Study report
Hospital
- Admission
- Scan
- Radiology report
- Discharge summary
Media
- Footage
- Script
- Captions
- Rights record
De-identification keeps those links intact: each person and patient gets one consistent pseudonym across every source.
How it works
From archive to licence in three steps.
Assess
Owners measure their data on their own systems. Only summary counts are shared, so we can match them to what labs are asking for.
Prepare
We index, clean, label and de-identify the data, working inside your systems wherever possible.
License & deliver
Datasets are built to each lab's spec and delivered with full chain of custody. You are paid on every licence.
Why Mullion
Quiet structure. Clear view.
A mullion is the slim bar that holds the glass in place. We do the same job for data: the part you don't see, holding everything together.
Rights first
Every dataset comes with a clear chain of title and licence terms labs can rely on.
Non-public
Data that isn't on the open web, so it adds signal models don't already have.
Built to spec
Pre-training, fine-tuning or evaluation sets, shaped to what each lab needs.
Owner control
Owners approve each buyer, each use and each term. Nothing is sold without a yes.