Agent risk alarms collide with just 39 safety openings

Security alarms around autonomous agents are getting louder, but hiring still steers away from safety. Our analyses this week found only 39 open roles in evaluation and safety, even as coverage dwelled on agent failures and cyber risk. One cut put those 39 inside a pool of 4,378 AI listings, reinforcing where budgets flow: modeling and infrastructure, not guardrails. Across the 198 employers we track, 4,160 AI roles were open as of 2026-08-30, down 83 on the week. Our model puts new AI jobs created month-to-date at roughly 170,639. The mix matters. If the loudest risks are not matched by dedicated hiring, the work of testing and hardening systems will sit with teams focused on building. That shapes career bets: candidates with safety skills face a thinner market today, while model and infrastructure talent still see the bulk of openings. It also sets expectations inside teams about who owns reliability when agents misbehave. The numbers do not settle the debate about how urgent the risks are. They do say where employers are placing near-term headcount, and for now that is not in safety.