Behaviorally fingerprinting Ox Alpha's provenance

Researchers analyzing "Ox Alpha," a new language model that appeared on OpenRouter in August 2026, determined it is part of Alibaba's GLM family through behavioral fingerprinting and tokenizer analysis. Unlike typical Chinese AI models that broadly censor foreign policy topics like Taiwan and Xinjiang, Ox Alpha exhibits selective censorship targeting only seven sensitive topics related to domestic politics and Xi Jinping, making it appear less censored by standard audits while actually employing a topic blacklist. The model's unique censorship pattern and linguistic markers suggest it represents a different approach to Chinese AI alignment focused on domestic political risk rather than broader geopolitical sensitivity.