For journalists & media partners

Polycreek is a Denver-based 501(c)(3) nonprofit that builds AI to catch online child grooming before it turns into abuse. Its model, Aletheia, spots the patterns of a grooming conversation while it's happening and flags which user is the predator. Trust and safety teams get back a scored list and start with the worst cases first.

Polycreek is a Denver-based 501(c)(3) nonprofit that uses machine learning to protect children from online exploitation. It was founded by software engineer Zeran Johannsen and works where machine learning meets trust and safety. Most tools only catch known illegal images after the harm is done. Polycreek focuses on grooming instead, the slow manipulation that happens in conversation before any abuse takes place. Its main model, Aletheia, is trained on more than 2.5 million conversations and scores direct messages, comments, and community chats in under 40 milliseconds. Each result comes back with a risk score, a confidence value, and the user who is most likely the predator. Aletheia doesn't act on its own. It ranks a review queue so trust and safety teams handle the highest-risk cases first. It also stores very little. Conversations aren't kept, and every result is timestamped so platforms can meet their obligations under COPPA, GDPR, DSA, and KOSA.

Online child exploitation is growing faster than human review can keep up with. In a single year, more than 20 million reports were sent to the U.S. National Center for Missing and Exploited Children. Existing tools are good at matching known illegal images, but they struggle to catch grooming, the slow manipulation in conversation that comes before abuse and leaves little evidence at the time.

Polycreek is a 501(c)(3) nonprofit working on that problem. It builds machine learning models that recognize the language and behavior of predatory contact. Its main model, Aletheia, is trained on more than 2.5 million conversations that come from public and private research datasets, real U.S. federal court records, and synthetic grooming scenarios. Aletheia works with messaging on social, gaming, and community platforms, reading conversations close to real time to catch grooming behavior and point to the user driving it.

The goal is to support people, not replace them. Every conversation gets a score from 0 to 1 and a rating from Safe to Critical, which sorts a review queue for trust and safety teams. Polycreek keeps no message text in storage, and it produces timestamped, review-ready records that platforms can use to escalate a case and to meet compliance requirements.