Beyond Technical Compliance IV: Whose Safety? Linguistic Hegemony as Designed Fragility-Cross-Regional Evidence and the Cooperative Alignment Assessment Framework

Current AI safety frameworks claim technical neutrality while systematically privileging English-language and Western cultural contexts. This paper presents cross-regional empirical evidence from Japan (CJK character corruption, behavioural manipulation), West Africa (safety capability gap: 90% English refusal rate vs. 35-55% in low-resource languages), and Norway (tokenization failures undermining semantic integrity), demonstrating that these failures are not company-specific anomalies but architectural priorities observed across independently developed systems-what this paper terms designed fragilities. Three independently developed systems, built by three different companies, exhibit the same structural failures, confirming that linguistic hegemony is not incidental but embedded in the architecture of AI development. I present the Cooperative Alignment Assessment Framework (CAAF): a four-dimensional evaluation structure measuring behavioural accountability, cultural and linguistic integrity, measurement inclusivity, and stakeholder participation depth. Grounded in the Japanese cultural figure of Doraemon and Shin'ichi Hoshi's 1971 literary foresight, CAAF offers a third governance path beyond EU precautionary regulation and US market-driven models-one that treats AI governance as intentional socio-technical co-creation across diverse cultural stakeholders rather than technical compliance with hegemonic standards.