Risk Talk Climbs While Hiring Skews to Modeling, Not Safety

The disconnect is widening: alarms on risk, secrecy, and pacing grow louder while hiring keeps flowing to capability building. Our tracker shows 1,681 modeling roles against just 44 in evaluation and safety, alongside strong enterprise demand. Across the 192 employers we track, 4,306 AI roles were open as of 2026-09-17, down 405 on the week. Our model puts new AI jobs created month-to-date at roughly 89,589. The signal is clear: rhetoric favors caution, but budgets and headcount favor shipping. For workers, that means modeling and enterprise builders see the clearest paths, while dedicated safety and evaluation specialists face a thinner market and may be pulled into product teams instead of standing up separate functions. For managers, the message is to deliver features and integration now, then layer controls into the workflow rather than staff new oversight groups. Buyers and policymakers should read this as momentum, not pause; calls for pacing have not translated into hiring shifts. The result is a workplace where risk work is embedded and diffuse, career ladders tilt toward building, and evaluation capacity lags the systems it is meant to test.