Reproducing Booz findings

A user attempted to reproduce findings from a Booz Allen Hamilton article claiming Chinese models are vulnerable. The user, who considers Qwen models among the best open-source options, is trying to manage risks while seeking optimal models. Contrary to Booz's claims, initial findings suggest Qwen 3 produced the best results among the four models tested. The user notes Booz's financial interest in deeming these models dangerous and finds their article and technical appendix difficult to reproduce for peer review.

This report offers a direct, albeit preliminary, counter-evidence to Booz Allen Hamilton's claims about Chinese model vulnerabilities, unlike other analyses that only cite the original report.