The Invisibility of Discard Through Acceleration: Governance of Generative AI Models (2024-2026)

The disappearance of generative AI models has shifted from an exceptional event to a recurringpattern, yet no proportional body of scholarship examines its social, epistemic, and politicalconsequences. Drawing on 24 months of observation (June 2024–June 2026), this paper documentstwo distinct mechanisms of invisibility in AI model governance. The first is commercial deprecation(n=6): in no analyzed case did official communication include a detailed technical specification of thespecific change, and 100% appealed to vague categories ("optimization," "scalability," "refocus"). Thesecond is state suspension: in June 2026, a US government export-control directive, invoking nationalsecurity without public evidence, forced the global suspension of Anthropic's Claude Fable 5 andMythos 5 models days after their launch; they were redeployed three weeks later without the groundsfor either decision being made public. Following Winner (1980), Rosa (2013), and De Angelis(2017), the paper argues that the combination of acceleration, technicality of justifications, and vagueappeals to security operates as a depoliticization mechanism that excludes from scrutiny not onlynon-technical users but—in the case of state suspension—even the affected company itself.