Abstract

We describe a system for detecting adult predatory behavior directed at minor users across five mobile messaging platforms (Discord, Instagram, Snapchat, SMS/MMS, Roblox) via on-device passive capture on the minor's own Android device. The system introduces three novel techniques not previously documented in either academic literature or commercial parental-monitoring products: (1) cross-platform stylometric fingerprinting that links blocked adult accounts to newly-created accounts by writing style rather than platform identifiers; (2) in-context grooming pedagogy delivering real-time named-tactic warnings to the child at the moment of manipulation, using non-punitive framing and one-tap protective action; and (3) three-stream ingestion with concealment-as-evidence semantics, where the deliberate absence of a message from an authoritative data source is treated as elevated forensic evidence of intent to conceal, and where classification tier upgrades trigger retroactive parent alerts. Each technique is described in sufficient detail to reproduce, establishing prior art. The combined system is designed for private family use, with a path to nonprofit release under an AGPLv3 + defensive-patent-pledge model.

Keywords: online child safety, grooming detection, stylometric fingerprinting, parental monitoring, on-device machine learning, chain of custody, NCMEC, CyberTipline, digital evidence

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.

Download

Share

COinS