WIRED reports that a combination of rapid capability advances, recursive self-improvement, and agentic swarms are genuinely alarming researchers inside major AI labs. The piece examines the convergence of technical factors driving existential risk concerns among frontier AI developers.