Most AI hiring builds. Little of it governs.
Role-anchored, only about 15–17% of the AI roles we've captured and tagged govern AI, by an LLM-coded ARISA role family not yet validated against human labels. A baseline, not yet a trend.
How much of AI hiring is actually governance
Across the covered firms, a minority of captured AI roles govern AI, and those that do cluster in a few role families.
Role-anchored, only about 15–17% of the AI roles we've captured and tagged govern AI. Counts include every captured posting, open or closed.
Cluster share by ARISA family across the tagged panel.
Each industry governs AI differently
The governance-hiring signature is sector-specific. No industry yet has enough tagged firms to report its own figure; each arrives as coverage lands, with its own regime.
- BankingToo few firms to report
Model Risk / SR 11-7, extended to AI
4 banks tagged — below the 8-firm floor, so no industry figure is reported
- Tech / AI-nativePartial
Frontier-lab safety, red-team, model security
build-heavy; governance thin (safety/security roles only)
- HealthcarePending coverage
Clinical-AI validation, FDA SaMD oversight, patient safety
coverage pending
- AutoPending coverage
Functional & AV safety (ISO 26262, UNECE R155/R156, SOTIF)
coverage pending
How the mix moves from here
You can track how this composition changes going forward. You cannot reconstruct the past from open postings on live ATS boards.
Live ATS boards show only what is hiring now, so past quarters cannot be rebuilt from them: the roles that opened and filled back then are already gone from the board (survivorship). Later points depend on future captures. The sibling “AI Hiring: US vs. EU” report uses a historical vendor panel instead; it is descriptive only and is not spliced onto this line.