Research

How we test Kibhi's safety, in the open.

Research papers by Tafadzwa Innocent Tauro, M.S. Cybersecurity, Quinnipiac University, measured on Kibhi itself. Every number comes from code we release with the papers.

Preprint, prepared for IEEE Transactions on Technology and Society

Refusals That Reveal: Server-Enforced Teen Contact Policies and Their Age-Inference Side Channels in a Production Social Platform

Tafadzwa Innocent Tauro · October 2026 · 7 pages · DOI 10.5281/zenodo.23147336

Protections that stop strangers contacting teens can, if built naively, reveal which accounts belong to teens. We found and closed two such leaks in Kibhi.

13 of 13 attack checks passed
Timing leak closed: 96% to chance
98.75% less phone-number probing

PDF Abstract and citation

Preprint, prepared for the Journal of Online Trust and Safety

LexObf-12: Measuring and Hardening the Obfuscation Robustness of a Production Lexical Content Filter

Tafadzwa Innocent Tauro · October 2026 · 5 pages · DOI 10.5281/zenodo.23147386

A benchmark of twelve ways people disguise banned words, and what it found in Kibhi's own filter: detection rose from 61.9% to 99.3% without more false alarms.

Detection 61.9% to 99.3%
Tested on 44,012 labeled posts
Open benchmark: LexObf-12

PDF Abstract and citation

Code and results

The LexObf-12 benchmark, the attack and timing test suites, and every measured result behind both papers.

Download the research package (.zip)

DOI 10.5281/zenodo.23147453

Author ORCID: 0009-0005-1693-2296