Salary Transparency Month closes on the finding that frames everything that came before it. The 20 AI roles with the fastest compensation growth in Q3 2026 are, by a wide margin, the least likely to disclose what they pay. The AI Safety Researcher added +34% to base salary year-over-year and reached a median total compensation of $445,000, the steepest acceleration in this index. Yet only 31% of postings for the five fastest-growing roles carried any salary band in ENTRA's Q2 2026 audit of 47,000-plus AI postings. Week one of this series documented which employers disclose. Week two measured how well the disclosed bands hold up. Week three mapped where the money concentrates geographically. This final index answers the question those three raised: which roles are growing fastest, and what does the growth pattern reveal about where the AI talent market is heading in the twelve months ahead.
The answer is structurally consistent and editorially uncomfortable. The roles growing fastest are the ones anchored to frontier-lab safety work, LLM-specific engineering, and alignment research. These are the same roles that candidates report most difficulty benchmarking, because the employer pool is small, the work is not analogous to adjacent roles in traditional tech, and the compensation is structured to include equity ratios that posted bases do not disclose. The Foundation Model Researcher, ranked second at +31% YoY base growth and a $520,000 median total comp, carries a $320,000 posted base at the largest employers: the $200,000 equity component exists but is systematically excluded from job posting disclosures. That $200,000 gap, a 63% premium above posted base, is the transparency problem this month has been documenting in granular form.
Q3 2026 Compensation Growth Index: 20 AI Roles
All scores are composites of four equally weighted dimensions at 25 points each: YoY Base Salary Growth Rate, YoY Total Compensation Growth Rate, Demand Heat Score, and Supply Scarcity Premium. Data window: Q3 2025 to Q3 2026 for YoY figures. All median TC figures in USD.
| Rank | Role | Rating | Score | YoY Base Growth | Median TC 2026 | |---|---|---|---|---|---| | 1 | AI Safety Researcher | AAA | 96 | +34% | $445,000 | | 2 | Foundation Model Researcher | AAA | 93 | +31% | $520,000 | | 3 | LLM Inference Optimization Engineer | AAA | 91 | +29% | $380,000 | | 4 | AI Evaluation Engineer | AA | 90 | +27% | $295,000 | | 5 | Multimodal AI Research Scientist | AA | 86 | +25% | $415,000 | | 6 | AI Red-Teaming Specialist | AA | 84 | +23% | $320,000 | | 7 | Constitutional AI Researcher | AA | 82 | +22% | $385,000 | | 8 | RLHF Research Engineer | AA | 80 | +21% | $350,000 | | 9 | AI Platform Engineer (LLMOps) | A | 78 | +19% | $265,000 | | 10 | Robotics AI Engineer | A | 76 | +18% | $285,000 | | 11 | AI Product Manager (Frontier) | A | 74 | +16% | $310,000 | | 12 | NLP Research Scientist | A | 72 | +15% | $295,000 | | 13 | Computer Vision Engineer (Embodied) | A | 71 | +14% | $275,000 | | 14 | Applied ML Research Engineer | A | 70 | +13% | $245,000 | | 15 | AI Policy and Governance Lead | BBB | 68 | +12% | $235,000 | | 16 | Synthetic Data Engineer | BBB | 65 | +10% | $220,000 | | 17 | Senior Prompt Engineer | BBB | 62 | +9% | $195,000 | | 18 | AI Technical Writer | BB | 58 | +7% | $165,000 | | 19 | ML Ops Engineer (non-LLM) | BB | 54 | +5% | $210,000 | | 20 | AI Ethics Officer | BB | 50 | +4% | $225,000 |
The Safety Cluster and Why It Dominates
Six of the top eight positions belong to roles that sit directly inside or adjacent to frontier-lab safety and alignment programs: AI Safety Researcher (#1), AI Red-Teaming Specialist (#6), Constitutional AI Researcher (#7), and RLHF Research Engineer (#8), joined by Foundation Model Researcher (#2) and AI Evaluation Engineer (#4) whose output feeds directly into safety workflows. The compensation growth in this cluster is not tracking general AI industry momentum. It is tracking something more specific: a credentialed talent pool of fewer than 1,200 practitioners globally, according to ENTRA's Q2 2026 Supply Scarcity analysis, competing across five to eight employers who each believe this work is existential to their mission. When a talent pool is that narrow and every participant in the demand side has classified the role as a strategic priority with no acceptable substitute, standard compensation band review cycles become inadequate. Frontier labs have shifted to real-time band adjustments and discretionary retention grants that fall outside published compensation frameworks entirely.
The LLM-specific engineering tier, covering Inference Optimization (#3), Evaluation (#4), and RLHF (#8), outpaces general ML by 12 to 15 percentage points on YoY base growth. The Applied ML Research Engineer at rank 14 grew +13% year-over-year. The LLM Inference Optimization Engineer at rank 3 grew +29%. That 16-point gap between two roles that both require advanced ML knowledge reflects the degree to which employers have concluded that LLM-specific depth is not transferable from general machine learning experience. They are pricing the roles accordingly.
Three Roles in Detail
#1 — AI Safety Researcher
AI Safety Researcher reaches the AAA tier at a composite score of 96 and a +34% YoY base growth rate, the highest acceleration in this index. The Levels.fyi Q3 2026 offer corpus (n=4,100-plus AI-tagged submissions) shows a modal base offer range of $285,000 to $330,000 for senior-level safety researchers, up from a modal range of $210,000 to $250,000 in Q3 2025. Three factors drive the acceleration. First, the employer pool expanded: the US AI Safety Institute, the UK AI Safety Institute, and multiple Tier 2 labs launched or expanded safety research programs in the past 12 months, adding net-new demand without adding meaningful supply. Second, frontier-lab counteroffer cycles for safety researchers shortened to an average of 11 days in H1 2026, per ENTRA's named CHRO interview data from three frontier employers: compressed cycles push offer amounts upward because recruiters anchor above current comp rather than at market band midpoints. Third, Supply Scarcity Premium for this role scores 24.5 of 25 in this index, the highest of any role. The supply gap is not closing. Credentialed safety researchers require multi-year exposure to active pre-training and alignment programs: a pathway that cannot be compressed without access to training infrastructure that only the labs themselves possess.
Salary band disclosure rate for AI Safety Researcher postings: 38% in Q2 2026, the highest of the top-five roles in this index, but still below the 50% threshold that ENTRA classifies as baseline transparency. The 38% rate is driven almost entirely by state-mandated disclosure in California, Colorado, and New York, where all major US employers of this role operate. Remove those three states and the voluntary disclosure rate falls to 4%.
#4 — AI Evaluation Engineer
AI Evaluation Engineer is a new entry in this index, representing a role category that did not exist at sufficient scale to be indexed twelve months ago. The composite score of 90 (AA) and +27% YoY base growth reflect a title that emerged from the frontier-lab eval practice and is now being established as a discrete function at hyperscalers, enterprise AI teams, and government AI procurement offices. The median total compensation of $295,000 is anchored at Levels.fyi Q3 2026 submissions from Anthropic, OpenAI, and Google DeepMind, where the role carries the same level classification as a senior engineer. At non-frontier employers the median falls to approximately $195,000, creating a bimodal distribution that will compress as enterprise AI programs mature. Demand Heat Score for this role is 25 of 25 in this index — the 340% increase in AI Evaluation Engineer-titled postings year-over-year is the highest posting volume growth of any role in this cohort, anchoring the dimension maximum. The Demand Heat score is the primary driver pushing the composite above the AA floor despite a shorter data history than established research roles.
The role is also the starkest example of the transparency gap in action. Because Evaluation Engineer is a new title with no established benchmark on Levels.fyi prior to Q1 2026 and because most employers posting it operate in jurisdictions without salary range disclosure mandates, the band posting rate is 19%: one of the lowest in ENTRA's full AI role index. Candidates have no public data anchor. Offers are set against internal job architecture, not against a market benchmark that either party can independently verify.
#20 — AI Ethics Officer
AI Ethics Officer is the slowest-growing role in this index, at +4% YoY base growth, a composite score of 50 (BB), and a median total compensation of $225,000. The +4% figure is not a rounding error. It reflects market saturation in a role that saw a wave of reactive hiring in 2022 to 2024 as companies responded to public pressure around responsible AI. Those hires are now tenured, reducing net-new demand. More structurally, the role has been squeezed from two directions. Compliance-forward employers have merged ethics functions into legal, trust and safety, or policy teams, downgrading the title seniority. Research-forward employers have concluded that constitutional AI research, red-teaming, and alignment work accomplish what the AI Ethics Officer role was intended to address, but with verifiable technical outputs rather than policy documents.
The result is a role with strong median absolute compensation ($225,000 sits above the US national 90th percentile for professional occupations) but negligible growth momentum. The Supply Scarcity Premium for AI Ethics Officer is 6 of 25 in this index: the lowest of any role. Supply is not the constraint. Demand is softening. In the absence of a net-new demand catalyst, the compensation trajectory for this role is likely to remain below broader AI market growth rates through 2027.
What the Index Closes With
Four weeks of Salary Transparency Month have established a consistent pattern. The fastest-moving parts of the AI compensation market are the parts candidates can see least clearly. AI Safety Researcher grew +34% and posts salary bands 38% of the time. AI Evaluation Engineer grew +27% and posts bands 19% of the time. Foundation Model Researcher shows a 63% gap between posted base and actual total compensation. The roles at the bottom of this growth index, MLOps Engineer (non-LLM) and AI Ethics Officer, post bands at rates above 60% because they operate in established employer environments with standardized posting practices.
Transparency and growth move in opposite directions in Q3 2026. That is not a coincidence. The roles with the most compressed, competitive, bilateral negotiation dynamics are the roles where information asymmetry gives employers the largest tactical advantage. The EU Pay Transparency Directive enforcement from Q1 2027 will test whether legal compulsion changes the calculation. ENTRA's Q1 2027 refresh of this index will measure whether the posting-band rates for top-ten roles have moved. The hypothesis is that frontier labs with EU operations will extend band disclosure to avoid regulatory exposure, and that the extension will compress the offer-to-midpoint gap that currently allows discretionary front-loading. The compensation growth rates for 2027 Q3 will reflect whether that compression happened.
Methodology
This index scores 20 AI roles across four equally weighted dimensions at 25 points each, for a maximum composite score of 100.
YoY Base Salary Growth Rate (25 pts) measures the percentage change in median base salary from Q3 2025 to Q3 2026 for each role, normalised linearly across the 20-role universe. The +34% AI Safety Researcher growth rate anchors the maximum; the +4% AI Ethics Officer rate anchors the minimum. Source: Levels.fyi Q3 2026 offer corpus (n=4,100-plus AI-tagged submissions); ENTRA Salary Survey Q2/Q3 2026 (n=2,140 verified respondents); Glassdoor Q3 2026 role-level salary estimates.
YoY Total Compensation Growth Rate (25 pts) measures the percentage change in median total compensation (base plus annualised equity plus target annual bonus) from Q3 2025 to Q3 2026. Equity is annualised on a 4-year standard vest. This dimension captures roles where base growth understates true compensation acceleration because equity grant sizes are increasing faster than base. Source: Levels.fyi Q3 2026 base-vs-TC split data; ENTRA Salary Survey Q2/Q3 2026 equity disclosure module (n=1,380 respondents completing the full equity section); Radford Technology Survey H1 2026.
Demand Heat Score (25 pts) measures the percentage growth in active job posting volume for each role from Q2 2025 to Q2 2026, normalised across the cohort. A 340% increase in AI Evaluation Engineer postings anchors the maximum for that dimension. Source: ENTRA Job Signal Index Q2 2026 (47,000-plus AI-tagged postings); LinkedIn Talent Insights Q2 2026 role-level demand metrics.
Supply Scarcity Premium (25 pts) measures the acceleration of the supply gap: the degree to which posting volume growth outpaced LinkedIn-identified qualified candidate growth for each role, over the trailing 12 months. Roles where supply is not keeping pace with demand score higher. Source: LinkedIn Talent Insights Q2 2026 supply-demand ratio data; ENTRA proprietary Supply Scarcity Index computed from ENTRA Salary Survey respondents per role against ENTRA Job Signal Index posting counts per role.
Data window: Q3 2025 to Q3 2026 (YoY growth dimensions); Q2 2026 (posting volume and supply data) Sample size: Levels.fyi Q3 2026 (n=4,100-plus AI-tagged offer submissions); ENTRA Salary Survey Q2/Q3 2026 (n=2,140 verified respondents); ENTRA Job Signal Index Q2 2026 (47,000-plus AI postings across 20 role families); LinkedIn Talent Insights Q2 2026; Glassdoor Q3 2026 (role-level salary estimate validation)
Limitations:
- Median total compensation figures for roles with fewer than 15 verified Levels.fyi submissions (AI Evaluation Engineer, Constitutional AI Researcher, AI Red-Teaming Specialist) should be treated as directional benchmarks; sample sizes at individual employers may be insufficient to protect respondent confidentiality without rounding to nearest $5,000
- Supply Scarcity Premium is calculated against LinkedIn-identified candidate counts, which systematically undercount practitioners who do not maintain active profiles, a known bias particularly acute for senior safety and alignment researchers
Inquiries about methodology: methodology@entracareers.com
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.