This page is a Google search results aggregation (not a primary source article). It collates claims and counterarguments about Dario Amodei's public statements warning that open-sourcing frontier AI models creates safety risks—specifically that once released, models can't be monitored, access can't be revoked, and dangerous capabilities can't be patched. The pro-centralization camp emphasizes dynamic safety controls; critics counter that this rhetoric functions as competitive gatekeeping that benefits well-capitalized firms, stifles transparency, and merely pushes capability development offshore. The source does cite real statements from Amodei in 2023 congressional testimony and recent media, but doesn't provide direct quotes or link to original testimony—only paraphrasing and reaction posts from Reddit, Instagram, and tech blogs. No new empirical data, test results, or policy analysis appears here. One sharp tension: if the geopolitical concern is real (that restricting US open-source just accelerates China's independent development), why hasn't that claim been stress-tested against actual capability timelines or economic factors?