Agent risk alarms grow as safety hiring stalls near 40 roles

Warnings about brittle agents and broader AI risks are getting louder, yet employers remain focused on shipping and scaling. Two analyses in our own data point to the gap: one counts only 39 open evaluation and safety roles, another just 40 spread across 19 firms, while most listings skew to builders and infrastructure. In the middle of that disconnect, the market keeps growing: across the 200 employers we track, 4,380 AI roles were open as of 2026-08-29, up 135 on the week, and our model puts new AI jobs created month-to-date at roughly 165,931. The picture for workers is uneven. If you work on safety and evaluation, demand looks thin and concentrated. If you build agents, platforms, or the plumbing that supports them, the hiring signal is strong, risks notwithstanding. For employers, the mix suggests a near term priority on capability over control. For now safety hiring is a rounding error relative to production builds. People choosing roles or reskilling should read it clearly: competence in agentic systems and the infrastructure around them is what companies are buying today, while safety expertise remains a smaller niche.