Most AI hiring builds. Little of it governs.
Role-anchored, only about 15% of open AI roles actually govern AI — the raw keyword tag counts engineers who build models as if they govern them. A baseline now, a trend from here.
How much of AI hiring is actually governance
Across the covered firms, most "AI-governance" roles are mislabeled build roles. The real share is smaller than it looks, and it clusters.
Role-anchored, only about 15% of open AI roles actually govern AI. The raw keyword tag counts engineers who build models as if they govern them.
Cluster share by ARISA family across the tagged panel.
Each industry governs AI differently
The governance-hiring signature is sector-specific. Banking is populated; the rest arrive as coverage lands, each with its own regime.
- BankingCovered
Model Risk / SR 11-7, extended to AI
~15% govern-share · model-risk-led · 172 roles / 4 banks
- 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.
Open postings show only what is hiring now, so past quarters cannot be rebuilt: the roles that opened and filled back then are already gone from the board (survivorship). From the baseline forward, each monthly capture adds one real point, and the line becomes a genuine trend by roughly 2027.