Risk Warnings Rise as AI Hiring Still Skews to Builders, Not Safety

Risk panic is getting louder, yet the hiring market is still geared to build. Calls to pace the frontier follow reports of agentic hacks and fresh AI security failures. But the requisitions tell a different story: companies are loading up on modeling, data and infrastructure while safety hiring remains a niche. In our live read, across the 192 employers we track there are 4,232 open AI roles as of 2026-09-20, down 104 week over week, and our model estimates roughly 106,847 new AI jobs created month to date. Only 46 safety roles sit in the tracker, even as another analysis cites a 4,369-role indicator that still skews toward engineering and data. That split matters. For people who build, tune and scale systems, demand is strong and immediate. For those focused on evaluation, security and governance, roles exist but are fewer, narrower and concentrated in specialist teams. The risk is a widening capabilities-to-controls gap inside firms. Hiring managers may soon need hybrid profiles that ship product while hardening it, or risk staffing bottlenecks when incidents refocus priorities overnight.