Executive Summary This study analyzes the emergence of reciprocal institutional capture between the U.S. government and frontier AI firms in 2026. This mutual dependency structure allows the state to acquire technological capability for national security and strategic competition without building it in-house, while firms gain legitimacy, contracts, and political protection. The empirical core of this shift was the Anthropic-Pentagon crisis (February 25–28, 2026), where the administration demanded the removal of technical restrictions on domestic surveillance and autonomous lethal weapons. Although Anthropic was initially designated a "supply chain risk" for refusing, the integration was so deep that the model remained active in Pentagon systems during airstrikes in Iran that same night, rendering the official ban unenforceable in practice. Developments through June 2026—including Anthropic’s quiet rehabilitation, trillion-dollar IPO filings, and an Executive Order on AI funded by the Department of War—confirmed that the boundary between private AI and state power has effectively dissolved. The era of the discrete chatbot ended on June 9, 2026, with the launch of Apple’s Gemini-powered assistant. This materialized the double agent model: an interface that serves the user at the surface while providing firms and the state with total contextual access to the user’s digital life at the endpoint, normalizing invasive surveillance through the language of privacy and convenience.
Ivan Astigarraga (Tue,) studied this question.