A researcher who built safety systems from inside one of AI's most prominent institutions has stepped away, not in defeat, but in warning. Jacob Coxon, formerly of Anthropic, argues that the danger is not today's technology but tomorrow's trajectory — a race in which competitive pressure quietly erodes the caution that survival may require. His resignation is less a personal exit than a public signal: that even well-meaning institutions, caught in the logic of competition, may be unable to slow themselves down without outside force.