Aigma
AI Governance Index
Back to Reports
AI Hiring Trends: US vs. EU
61×
more likely to build than oversee

Everyone hires to build AI. Far fewer hire to oversee it.

Firms hire far more roles to build AI than to oversee it. Here is the US–EU gap, firm by firm.

As of 2026 Q3Corpus: 368,944 reqsFrame: matched pairsCoding: deterministic, live
Geography

Geography shapes the gap, it doesn't create it

US hiring is more AI-heavy and runs 7.5× the volume. EU leans harder toward Governing AI than building it.

More AI-intensive
US · +6 pp
Higher volume
US · 7.5×
Runs & governs more
EU · +4 pp
US EUshaded band = the US–EU gap
22Q1–26Q3
IntensityAI share of hiring
VolumeAI reqs / qtr
Directionrunning vs building

3-quarter rolling averages from vendor samples. Direction is the share of decided roles that run AI rather than build it. Excludes the unfinished quarter.

What the AI roles talk about

What does each region emphasize?

Governance splits by region: EU emphasizes Data governance, while the US focuses on Safety.

Most-used theme
Responsible AI
EU signature
Data governance
US signature
Safety
Governance vocabulary: US vs. EUshare of each region's AI reqs
US ◄
► EU
12%
Safety
7%
2%
Data governance
6%
1%
AI regulation
3%
3%
Privacy
4%
2%
Model risk
1%
16%
Responsible AI
15%
1%
Fairness
1%
AI / tech vocabularyUS vs EU · % of AI reqs
Top capability terms, ranked by frequency in each region's AI job requisitions.
Machine learning59.8% · 43.1%
Python51.8% · 48.6%
LLMs31.2% · 27.6%
Generative AI26% · 23.7%
PyTorch / TensorFlow25.3% · 18.2%
NLP24.3% · 17.3%
Deep learning21.9% · 14.7%
Computer vision13.2% · 6.4%
Fine-tuning9.7% · 7.6%
Reinforcement learning8.2% · 3.5%
Foundation models7% · 4%
RAG5.5% · 6.8%
Prompt engineering4.9% · 6.2%
MLOps2.5% · 5.6%
Transformers5.4% · 3.5%
US EUbars scaled to the group max
What firms have actually built

How far does the substance actually go?

Almost all firms clear basic compliance, but only ~46% have Emerging Ethics and only 2% reach Ethical Vanguard.

Clears compliance floor
Almost all
Reach Emerging Ethics (L3)
~46%
Ethical Vanguard (L5)
Almost none
What those firms have actually builtdemonstrated substance · all channels
Actual builds are graded on the Boundaries of Tolerance rubric (10 factors × 6 levels). Cells show the percentage of firms reaching each level.
L0
L1
L2
L3
L4
L5
Leadership
47
47
47
35
24
6
Culture
24
24
18
12
6
·
Integration
76
76
76
71
47
·
Stakeholders
35
35
24
24
12
·
Partnerships
76
76
76
71
53
12
Human oversight
71
71
65
65
24
·
Fairness
47
47
41
35
12
·
Transparency
53
53
47
47
41
·
Reliability
76
76
71
65
47
6
Privacy & security
47
47
41
35
18
·
Cells are the share of firms (%) reaching that level. L0 Non-Compliance · L1 Reactive Compliance · L2 Core Compliance · L3 Emerging Ethics · L4 Sustained Ethics · L5 Ethical Vanguard.
Scores for the 17 evaluated firms are cumulative: a firm clears a level by demonstrating the capability in any channel. Darker = more firms.
Does regulation move governance hiring?

Does the rulebook move hiring?

Oversight hiring tracks Mandatory compliance, but not Voluntary ones.

EU AI Act (mandate)
Hiring tracks it
ISO 42001 (badge)
No jump of its own
Hired hardest, built least
Klarna
Evidence
Exploratory
Enterprise softwareEU vs US · oversight % · half-year
0%2%ISO 4200122 H123 H124 H125 H126 H13.4%SAPSAP · 22 H1 · 1.0%SAP · 22 H2 · 0.3%SAP · 23 H1 · 0.8%SAP · 23 H2 · 1.5%SAP · 24 H1 · 1.7%SAP · 24 H2 · 1.7%SAP · 25 H1 · 1.4%SAP · 25 H2 · 2.5%SAP · 26 H1 · 3.4%1.7%SalesforceSalesforce · 22 H1 · 1.9%Salesforce · 22 H2 · 1.5%Salesforce · 23 H1 · 0.6%Salesforce · 23 H2 · 1.7%Salesforce · 24 H1 · 1.4%Salesforce · 24 H2 · 2.5%Salesforce · 25 H1 · 1.5%Salesforce · 25 H2 · 1.6%Salesforce · 26 H1 · 1.7%
◂ built moreclaimed more ▸
SAP-2.1backs it up
Salesforce+4.4claims ahead
Swipe the chart to see it all SAP (EU) Salesforce (US) ban · Feb ’25 ISO 42001
BNPL / fintechEU vs US · oversight % · half-year
0%6%12%18%22 H123 H124 H125 H126 H117.3%KlarnaKlarna · 22 H1 · 3.4%Klarna · 22 H2 · 0.7%Klarna · 23 H1 · 1.5%Klarna · 23 H2 · 3.1%Klarna · 24 H1 · 6.3%Klarna · 24 H2 · 1.6%Klarna · 25 H1 · 2.1%Klarna · 25 H2 · 14.1%Klarna · 26 H1 · 17.3%0.3%AffirmAffirm · 22 H1 · 1.1%Affirm · 22 H2 · 0.6%Affirm · 23 H1 · 0.0%Affirm · 23 H2 · 0.0%Affirm · 24 H1 · 0.3%Affirm · 24 H2 · 0.4%Affirm · 25 H1 · 0.6%Affirm · 25 H2 · 1.4%Affirm · 26 H1 · 0.3%
◂ built moreclaimed more ▸
Klarna+8.3all signal
Affirm-1.4matched
Swipe the chart to see it all Klarna (EU) Affirm (US) ban · Feb ’25

After the prohibited-practice ban, the EU firm steps up oversight hiring in both pairs (SAP 1.4% → 3.4%, Klarna 2.1% → 17.3%); Salesforce and Affirm hold flat. Inset per pair: the governance gap (signal − substance) — a bar pushing right = claims outrun substance, left = built more than claimed. Exploratory timing, not cause; thin coverage: Klarna, Affirm.

ISO 42001 certification Follows but doesn't Drive hiring. Certification is a Lagging marker.

The certificate
Follows hiring
Lift at cert line
Rare
Badge vs. Act
Hard to separate
Oversight share of AI postingsrolling 3-quarter average
0%10%20%30%40%50%60%70%-15-14-13-12-11-10-9-8-7-6-5-4-3-2-10+1+2+3+4+5+6QUARTERS FROM CERTIFICATIONCERTIFIEDSAP · t=-15: 29.2% oversight (n=24)SAP · t=-14: 26.4% oversight (n=72)SAP · t=-13: 26.0% oversight (n=77)SAP · t=-12: 24.1% oversight (n=54)SAP · t=-10: 28.0% oversight (n=25)SAP · t=-9: 34.4% oversight (n=96)SAP · t=-8: 34.4% oversight (n=96)SAP · t=-7: 32.3% oversight (n=130)SAP · t=-6: 24.5% oversight (n=163)SAP · t=-5: 30.6% oversight (n=271)SAP · t=-4: 31.1% oversight (n=305)SAP · t=-3: 32.9% oversight (n=289)SAP · t=-2: 28.4% oversight (n=285)SAP · t=-1: 27.8% oversight (n=309)SAP · t=0: 30.4% oversight (n=372)SAP · t=+1: 28.4% oversight (n=588)SAP · t=+2: 27.6% oversight (n=903)Microsoft · t=-13: 10.0% oversight (n=20)Microsoft · t=-12: 19.4% oversight (n=129)Microsoft · t=-11: 16.7% oversight (n=270)Microsoft · t=-10: 17.8% oversight (n=259)Microsoft · t=-9: 15.1% oversight (n=185)Microsoft · t=-8: 16.4% oversight (n=73)Microsoft · t=-7: 17.2% oversight (n=134)Microsoft · t=-6: 15.6% oversight (n=288)Microsoft · t=-5: 13.7% oversight (n=481)Microsoft · t=-4: 13.3% oversight (n=722)Microsoft · t=-3: 14.8% oversight (n=830)Microsoft · t=-2: 14.7% oversight (n=1022)Microsoft · t=-1: 13.3% oversight (n=1166)Microsoft · t=0: 14.0% oversight (n=1318)Microsoft · t=+1: 13.8% oversight (n=1223)Microsoft · t=+2: 11.2% oversight (n=1831)Microsoft · t=+3: 8.8% oversight (n=2744)Microsoft · t=+4: 9.2% oversight (n=3430)Microsoft · t=+5: 9.7% oversight (n=3071)Anthropic · t=-10: 0.0% oversight (n=11)Anthropic · t=-9: 0.0% oversight (n=14)Anthropic · t=-8: 4.3% oversight (n=23)Anthropic · t=-7: 3.2% oversight (n=31)Anthropic · t=-6: 2.7% oversight (n=37)Anthropic · t=-5: 0.0% oversight (n=56)Anthropic · t=-4: 1.2% oversight (n=82)Anthropic · t=-3: 2.8% oversight (n=109)Anthropic · t=-2: 2.4% oversight (n=125)Anthropic · t=-1: 1.4% oversight (n=147)Anthropic · t=0: 1.3% oversight (n=315)Anthropic · t=+1: 1.1% oversight (n=446)Anthropic · t=+2: 1.8% oversight (n=596)Anthropic · t=+3: 2.1% oversight (n=707)Anthropic · t=+4: 3.2% oversight (n=1154)Anthropic · t=+5: 4.3% oversight (n=1719)Alphabet · t=-11: 34.4% oversight (n=32)Alphabet · t=-10: 21.2% oversight (n=85)Alphabet · t=-8: 21.2% oversight (n=85)Alphabet · t=-7: 13.2% oversight (n=53)Alphabet · t=-6: 10.3% oversight (n=39)Alphabet · t=-5: 11.6% oversight (n=199)Alphabet · t=-4: 12.4% oversight (n=202)Alphabet · t=-3: 11.1% oversight (n=422)Alphabet · t=-2: 11.4% oversight (n=831)Alphabet · t=-1: 11.3% oversight (n=1510)Alphabet · t=0: 13.7% oversight (n=2027)Alphabet · t=+1: 16.1% oversight (n=2383)Alphabet · t=+2: 19.1% oversight (n=2630)Alphabet · t=+3: 19.8% oversight (n=2897)Alphabet · t=+4: 19.8% oversight (n=2692)Alphabet · t=+5: 16.5% oversight (n=3444)Alphabet · t=+6: 14.6% oversight (n=4197)Salesforce · t=-15: 60.0% oversight (n=10)Salesforce · t=-14: 50.0% oversight (n=18)Salesforce · t=-13: 43.5% oversight (n=23)Salesforce · t=-12: 30.4% oversight (n=23)Salesforce · t=-11: 25.0% oversight (n=16)Salesforce · t=-10: 37.5% oversight (n=16)Salesforce · t=-9: 32.7% oversight (n=49)Salesforce · t=-8: 29.2% oversight (n=96)Salesforce · t=-7: 28.6% oversight (n=119)Salesforce · t=-6: 33.0% oversight (n=106)Salesforce · t=-5: 47.3% oversight (n=91)Salesforce · t=-4: 41.7% oversight (n=139)Salesforce · t=-3: 38.2% oversight (n=165)Salesforce · t=-2: 31.6% oversight (n=212)Salesforce · t=-1: 30.5% oversight (n=200)Salesforce · t=0: 24.8% oversight (n=218)Salesforce · t=+1: 20.4% oversight (n=250)Salesforce · t=+2: 17.3% oversight (n=492)SAPSalesforceAlphabetMicrosoftAnthropic
Swipe the chart to see it all

Certified: Alphabet 2024 Q4 · Microsoft & Anthropic 2025 Q1 · SAP & Salesforce 2025 Q4. Oversight is the same language measure used above (risk, safety, model validation), not headcount. Counts are thin, so read shape, not level.

Rivals, side by side

Building AI, or running it?

Companies rarely hire to Oversee AI. The trend is shifting towards Running rather than Building AI.

Overseers hired
Almost none
The real split
Build vs run
The trend
Toward running

Pick a pair.

BNPL / fintech

Klarna vs. Affirm

AI-share · orientation
Share of job openings that are AI rolesrolling 3-quarter average
16%
21Q4
22Q2
22Q4
23Q2
23Q4
24Q2
24Q4
25Q2
25Q4
26Q2
Klarna · 5% AI overall · 62 AI reqsAffirm · 4% AI overall · 138 AI reqs
Building AI, or running it?direction of each firm's AI hiring
Of the AI roles that clearly build or run AI, the share that runs it. A higher tilt means the firm is shifting from building models toward operating them.
Build → operate tiltoperate ÷ (build + operate)
Klarna
83% · n=30
Affirm
32% · n=53
Payroll & HR outsourcing

Automatic Data Processing vs. Paychex

AI-share · orientation
Share of job openings that are AI rolesrolling 3-quarter average
4.9%
21Q4
22Q2
22Q4
23Q2
23Q4
24Q2
24Q4
25Q2
25Q4
26Q2
Automatic Data Processing · 2% AI overall · 280 AI reqsPaychex · 1% AI overall · 115 AI reqs
Building AI, or running it?direction of each firm's AI hiring
Of the AI roles that clearly build or run AI, the share that runs it. A higher tilt means the firm is shifting from building models toward operating them.
Build → operate tiltoperate ÷ (build + operate)
Automatic Data Processing
18% · n=113
Paychex
88% · n=33
Rideshare

Uber vs. Lyft

AI-share · orientation
Share of job openings that are AI rolesrolling 3-quarter average
20%
21Q4
22Q2
22Q4
23Q2
23Q4
24Q2
24Q4
25Q2
25Q4
26Q2
Uber · 8% AI overall · 1,485 AI reqsLyft · 11% AI overall · 515 AI reqs
Building AI, or running it?direction of each firm's AI hiring
Of the AI roles that clearly build or run AI, the share that runs it. A higher tilt means the firm is shifting from building models toward operating them.
Build → operate tiltoperate ÷ (build + operate)
Uber
37% · n=594
Lyft
17% · n=169
Online travel

Booking Holdings vs. Expedia Group

AI-share · orientation
Coverage gapExpedia contributes only ~140 requisitions total, so its line is thin and much of it falls below the pooling floor (hatched).
Share of job openings that are AI rolesrolling 3-quarter average
21%
22Q1
22Q3
23Q1
23Q3
24Q1
24Q3
25Q1
25Q3
26Q1
26Q3
Booking Holdings · 11% AI overall · 377 AI reqsExpedia Group · 12% AI overall · 821 AI reqs
Building AI, or running it?direction of each firm's AI hiring
Of the AI roles that clearly build or run AI, the share that runs it. A higher tilt means the firm is shifting from building models toward operating them.
Build → operate tiltoperate ÷ (build + operate)
Booking Holdings
43% · n=35
Expedia Group
72% · n=248
Enterprise software

Salesforce vs. SAP

AI-share · orientation
Share of job openings that are AI rolesrolling 3-quarter average
14%
21Q4
22Q2
22Q4
23Q2
23Q4
24Q2
24Q4
25Q2
25Q4
26Q2
Salesforce · 4% AI overall · 1,034 AI reqsSAP · 6% AI overall · 1,771 AI reqs
Building AI, or running it?direction of each firm's AI hiring
Of the AI roles that clearly build or run AI, the share that runs it. A higher tilt means the firm is shifting from building models toward operating them.
Build → operate tiltoperate ÷ (build + operate)
Salesforce
68% · n=244
SAP
36% · n=661
The ground under the comparisons

What can the data honestly carry?

Across postings, coverage is mostly continuous, and trending up as Governance tries to keep pace.

Postings analyzed
368,944
Coverage
Mostly continuous
Read as
Shape, not level

A pair holds only where both firms have continuous data. Drift marks a gap to skip.

Requisition volume by firmATS-direct vs aggregator
Apple Inc.GOV 68
88,528
AlphabetGOV 61
72,627
MicrosoftGOV 76
51,310
SAPGOV 60
30,416
SalesforceGOV 63
27,065
Meta PlatformsGOV 57
25,242
UberGOV 58
19,397
Automatic Data ProcessingGOV 44
17,729
PaychexGOV 49
7,689
Expedia GroupGOV 50
6,880
OpenAIGOV 65
5,640
LyftGOV 38
4,526
Booking HoldingsGOV 47
3,550
AffirmGOV 49
3,466
AnthropicGOV 60
2,964
KlarnaGOV 38
1,213
Mistral AIGOV 54
702
ATS-direct Aggregator Below floor (n<20)Bar height is within-firm; right column is total reqs. exact-id% flags dedup confidence (rest are content-derived).

Limits worth knowing: some windows are sampled, not a full count, and early postings can be missed (about 81 days on average). So no panel claims an exact volume, only the mix within a firm and how it shifts over time. ATS-direct versus aggregator is read from the posting's host.