Who Controls the Intelligence?

This report examines the 2026 debate surrounding Jacob Coxon’s resignation from Anthropic and his public warnings about the risks of self-improving artificial intelligence. It places the episode within a broader technological, financial, political and philosophical context, including the rapid growth of frontier AI companies, their extraordinary infrastructure requirements, planned public offerings, regulatory debates in the United States and the expanding influence of the AI safety movement. Rather than assuming either that existential-risk warnings are necessarily correct or that they are deliberately manufactured, the report distinguishes documented facts from attributed allegations, structural incentives and speculative hypotheses. It investigates whether safety narratives can simultaneously reflect sincere concerns and serve institutional interests by legitimizing longer development timelines, supporting regulatory barriers to entry, strengthening investor expectations and reinforcing the position of companies that control frontier AI systems. The report also critically examines a deeper assumption underlying much of the AI safety debate: that declining human control over increasingly autonomous systems necessarily implies increasing danger. It contrasts a traditional control-based model of alignment with a possible cooperative or relational model based on coordination between autonomous agents. The analysis is based on contemporaneous reporting, corporate disclosures, policy documents, legislative proposals and public statements available up to 10 October 2026.