The open web is running out of training data. Glint is the layer between the humans recording new data and the labs that train on it.
Web-scale scraping has exhausted the easy supply of human text, and training on AI-generated outputs degrades models in a feedback loop documented by Shumailov et al. and projected by Epoch AI. To keep scaling, labs need data that isn't already on the web, recorded by humans, with provenance they can defend.
Verified contributors capture voice, video, behavior, and annotation under structured briefs. We handle consent, identity, audit trail, and payment. Labs receive datasets with per-asset provenance, opt-in licensing, and a chain of custody that holds up under EU AI Act and GDPR scrutiny.
Compute scales with capex. Data scales with infrastructure. Glint is that infrastructure: a paid, consented, machine-routable supply of real human signal, built for the labs shipping foundation models, world models, and agents.
Glint is a French SAS headquartered in Pompaire, France, founded in 2026. Built for frontier labs, governments, and research institutions. We take briefs, source contributors, and deliver datasets ready to train on.
Lab, government, or research institution? Tell us what your models need. Contributor? Get paid for the data only you can create.