Last Friday, this publication quantified the pay transparency dividend: AI engineers who negotiate against a posted salary band capture an average of $43,000 more in year-one total compensation than those negotiating blind, per ENTRA Q2 2026 Salary Survey data across 2,140 practitioners globally. The conclusion was that band posting generates measurable returns for engineers and employers alike, and that the economic case for disclosure has now surpassed the regulatory case in persuasive force.
This report takes the next step, and it is an uncomfortable one for transparency advocates who treat band posting as the finish line.
California SB 1162's first annual pay data cycle closed in May 2026, producing the first large-scale mandatory dataset revealing what companies actually pay across gender lines within the bands they post. The findings are not what the law's authors intended to demonstrate. Across the technology sector employers captured in the California Civil Rights Department 2025 Pay Data Report, the mean unadjusted gender pay gap in the professional computing occupations category was 17.4 percent. The majority of those employers post salary ranges. Most are legally compliant under state and local disclosure mandates. The gap persists not because companies are hiding the band. It persists because the band, as a disclosure instrument, was never designed to reveal who inside it earns what.
This is the second-order problem of salary transparency: the legal compliance layer and the equity layer are not the same layer, and the current generation of disclosure frameworks conflates them. A company can post a band of $180,000 to $280,000 for a Senior ML Engineer role, achieve 100 percent posting compliance across all mandated jurisdictions, and simultaneously have women concentrated at $183,000 to $215,000 and men clustered at $248,000 to $280,000 within that identical range. The band is transparent. The distribution is not. California's 2025 pay data cycle is the first systematic documentation that this gap exists at sector scale.
The band-disclosure regime has three structural blind spots that the current Salary Transparency Month conversation has underweighted. The first is gender composition: which employees within the band are concentrated at the floor versus the ceiling, and why. The second is career-stage compression: how cross-level band architecture conceals the practical compensation gap between entry-level and mid-career practitioners holding identical job titles. The third is geographic arbitrage: how remote-first "local market rate" policies create within-team pay gaps that the same posted band does not address or disclose.
Each of these dimensions has quantitative evidence behind it. None of them can be closed by posting a number. That is the thesis of this report, and the data behind it is what Salary Transparency Month 2026 has produced.
1. The Gender Composition Problem Inside the Band
California SB 1162, enacted in September 2022 and in full effect for employer filings beginning with calendar year 2023, was the most consequential US pay transparency legislation in a generation for one reason that distinguishes it from the posting mandates active in Colorado, New York City, and Washington: it requires employers to report actual pay data by gender and race, disaggregated by job category, not just to post a range on a listing. The first mandatory annual filing cycle covering calendar year 2025 data closed May 13, 2026. The California Civil Rights Department published aggregate sector-level findings in June 2026.
The technology sector figures from the California CRD 2025 Pay Data Report are precise and damaging in their precision. Across professional computing occupations at California-covered employers with 100 or more employees, women represented 28.4 percent of the filing population but 19.1 percent of employees in the highest pay band reported within each employer's own internal grade structure. The unadjusted mean gender pay gap across the sector was 17.4 percent. Controlled for years of experience and job title at the employer level, the CRD's adjusted gap estimate was 6.2 percent. Both numbers matter. The unadjusted 17.4 percent reflects labor market structural conditions including occupational segregation; the adjusted 6.2 percent reflects what employers pay differently, within identical role and experience categories, on the basis of gender. Neither figure would be visible from reading job postings. Both are consequences of within-band pay distribution that band disclosure does not surface.
The company-level data from annual gender pay reporting adds specificity. UK regulations require employers with 250 or more UK employees to publish mean and median gender pay gap data annually, producing a cross-company comparison set that does not exist in the United States at company level.
| Company | Jurisdiction | Mean gender pay gap | Source | |---|---|---:|---| | Amazon | United Kingdom | 18.0% | Amazon UK Gender Pay Report 2025 | | Meta | United Kingdom | 16.4% | Meta UK Gender Pay Report 2025 | | Microsoft | United Kingdom | 12.8% | Microsoft UK Gender Pay Report 2025 |
Mean gap defined as (male mean pay minus female mean pay) / male mean pay, per UK statutory methodology.
These figures exist alongside full compliance with UK salary disclosure practices at several of the same companies. Google DeepMind posts salary ranges on a majority of its UK research roles. Microsoft UK's disclosure rate across AI and engineering roles exceeds 60 percent per ENTRA Pay Transparency Audit Q2 2026 monitoring. The gap and the disclosure coexist because they operate at different analytical levels: disclosure reveals the range, not the distribution within it.
The Structural Cause: Who Holds Top-of-Band Roles
The mechanism producing within-band gender concentration is not band-level behavior. It is role-category segregation at the level below the band, within roles that share a title and a posted range.
McKinsey and LeanIn.org's "Women in the Workplace 2025" report, published September 2025, found that women held 31 percent of senior individual contributor roles in the technology sector broadly and 22 percent of senior IC roles in what the study classified as "deep technical" specializations, which include model research, infrastructure optimization, and compute efficiency engineering. Those are precisely the specializations concentrated at the top of AI compensation bands. An L5 or L6 Foundation Model Researcher at a frontier lab earns total compensation 40 to 70 percent above the base salary posted in the band. The base band is compliant. The equity composition of the population earning above the band's midpoint is not disclosed and, under current frameworks, is not required to be.
The American Association of University Women's 2026 STEM pay analysis found that women in computing and mathematical occupations in the United States earn approximately 83 cents per dollar earned by men in equivalent roles, a gap that narrows to approximately 91 cents when controlling for hours worked, years of experience, and educational attainment but does not close. The residual 9-cent gap at equivalent credentials and experience represents the within-band distribution effect that SB 1162's pay data reporting is designed to document and, over successive cycles of enforcement, to reduce.
The equity problem is structural because the roles that command top-of-band pay are not randomly distributed across gender lines. ENTRA's Q2 2026 Pay Transparency Audit found that across 87 companies in the audit cohort, Foundation Model Research roles were held by women at a rate of approximately 18 percent. Inference Optimization Engineering roles: 21 percent. Applied Scientist roles at the mid-market applied-AI tier: 34 percent. The within-band concentration of women toward the floor of senior AI compensation bands is a reflection of the distribution of women across the role sub-categories that produce top-of-band and above-band total compensation, not exclusively a reflection of within-role pay discrimination. Closing the within-band gap requires addressing both dimensions, and posting a number addresses neither.
2. Career-Stage Compression: The Architecture of Disadvantage
The second hidden dimension of pay inequity is less visible than the gender gap and more directly caused by the design choices of the employers who structure it. It operates at the intersection of two facts that are both individually transparent but jointly opaque: frontier labs post career-spanning salary bands, and those bands are presented to early-career candidates as if the entire range is available to someone beginning their career.
Anthropic's Senior Researcher and Research Engineer band, posted on roles across its San Francisco, New York, and Seattle listings, runs from $120,000 to $320,000 base salary. That range spans the company's L3 through L5 equivalent levels on the research ladder: from a new PhD entering the research track to a senior researcher with eight or more years of experience and a track record of independent contributions to frontier model development. The band is legally compliant under California, New York, and Colorado disclosure laws. It covers every possible placement within three distinct career stages. A new PhD entering the research track at Anthropic is presented with a posting that lists "$320,000" as the ceiling of the range for their role.
The Levels.fyi composite for PhD-direct AI research offers at frontier labs in Q2 2026 shows that entry-level research scientist hires at Anthropic, OpenAI, and Google DeepMind typically receive base salary offers in the range of $200,000 to $280,000, with total compensation including equity and signing bonus in the $280,000 to $420,000 range. The $120,000 floor in Anthropic's posting is not a number that corresponds to any offer Anthropic has made to a PhD-direct research hire in the recent hiring cycle. It is the floor of the range because it represents the minimum base across the full three-level band. A candidate who opens their negotiation at $190,000 because the floor is $120,000 and they want to be "reasonable" has cost themselves approximately $80,000 to $90,000 in base salary before the first conversation.
The Survey Evidence: 34 Percent and $78,000 to $95,000
ENTRA's Q2 2026 Salary Survey (n=2,140) captured compensation data with career-stage granularity sufficient to quantify the cross-level compression effect. Respondents were classified into three experience cohorts: early-career (fewer than three years of professional AI experience post-degree), mid-career (three to eight years), and senior (more than eight years). Within each equivalent job title, the early-career cohort's reported offer final compensation was compared against the mid-career cohort's for the same title and employer tier.
The finding: early-career AI practitioners in the ENTRA Q2 2026 cohort averaged 34 percent below mid-career practitioners in equivalent-titled roles at equivalent employer tiers. Expressed in absolute terms, the gap averaged $78,000 annually at the applied-AI tier (AI-native companies outside the frontier-lab cohort) and $95,000 annually at the frontier-lab tier, where absolute compensation levels are higher and the cross-level gap is therefore wider in dollar terms. Those gaps appear, in job postings, as identical ranges. An Anthropic Senior Researcher posting carrying a $120,000 to $320,000 band applies to both the early-career respondent who received a $195,000 base and the mid-career respondent who received a $290,000 base. Both are "within the posted band." The band conceals a $95,000 difference that the posting was specifically designed to contain.
The anchoring consequence is directional and well-documented in behavioral economics literature. When candidates encounter a wide range with a visible floor, the floor exerts more anchoring influence on the candidate's opening offer than the ceiling does, particularly for candidates who lack third-party benchmark data or prior-cycle experience negotiating equivalent roles. ENTRA Q2 2026 survey respondents in the early-career cohort who reported opening at the band midpoint or above captured final offers averaging 12 percent higher than early-career respondents who opened at or below the midpoint. The difference in opening strategy translated to a $23,000 to $28,000 average difference in final offer at the applied-AI tier. The posted band was identical for both groups. The behavioral response to the floor was not.
The Wide-Band Design as Policy Instrument
The $200,000 band spread at Anthropic, and equivalent spreads at OpenAI ($100,000 to $300,000 plus additional compensation) and Google DeepMind (which does not publish a single global band but whose US filings under state disclosure law show research engineer spreads exceeding $160,000), is not a consequence of compensation complexity. It is a design choice with a specific function: containing multiple career levels within a single posted range, which satisfies disclosure law while preventing candidates from locating their level-appropriate compensation without inside information.
Germany's Entgelttransparenzgesetz, enacted July 2017, with individual pay information rights effective January 2018, provides the regulatory model for addressing this. The EntgTranspG requires that when an employee submits a pay transparency request, the employer must provide the median compensation of at least six employees in a "comparable group" and explain the criteria by which the group was defined. The EU Pay Transparency Directive 2023/970/EU, whose transposition Germany is working toward with projected implementing legislation in early 2027, goes further: it requires that salary ranges in job postings be based on "objective, gender-neutral criteria" and that the range disclosed corresponds to the actual pay range applicable to the specific role, not an aggregate band spanning multiple career levels.
The practical implication of the EU model, when it applies to AI employers with EU operations, is that a single $200,000-spread band posted for a research role spanning L3 through L5 would fail the "objective criteria" test. Level-specific ranges, or at minimum seniority-bounded ranges, would be required. Stripe, which publishes compensation bands broken out by level on its public careers page for many engineering roles, represents the voluntary implementation of this model in the current US regulatory environment. Stripe's disclosed senior engineer base range for its San Francisco office runs approximately $210,000 to $240,000 for its senior L5-equivalent level, a spread of $30,000 rather than $200,000. At that width, early-career candidates cannot reasonably be misled about where they fall in the band, because the band does not span eight years of career progression.
3. Geographic Arbitrage: Same Band, Unequal Pay
The third hidden dimension is the newest and, in quantitative terms, potentially the largest. The global expansion of remote-first and remote-eligible hiring at AI companies has created a compensation structure in which the same job title, the same posting, and in many cases the same published band applies to employees whose actual cash compensation differs by 40 to 60 percent based solely on their geographic location. This is legal, disclosed (in the sense that companies publish their geographic compensation philosophy), and invisible in the mechanism that the current pay transparency framework relies upon: the posted salary range.
Hugging Face's compensation philosophy, published on its careers page, states that base salary is set by reference to local market rates in the currency of the employee's country of employment. This is the standard "local market rate" approach adopted by the majority of global AI employers. The consequences for within-team pay equity are visible in the company's own job listings. A ML Research Engineer role posted in Paris carries a base salary range of €108,000 to €142,000. The equivalent ML Research Engineer role posted for San Francisco carries a base salary range of $210,000 to $280,000. The roles share a title, a level, and at Hugging Face, a compensation framework described in the same public document. The cash compensation gap between a Paris-based researcher at midpoint (€125,000, approximately $136,000 at the canonical EUR/USD rate of 1.09) and a San Francisco-based researcher at midpoint ($245,000) is 44 percent at band midpoints and reaches 48 percent when computed against ENTRA Q2 2026 Remote Work Survey data using actual offer distributions rather than posted band midpoints.
The Within-Team Pay Architecture
The geographic gap is not a comparison between two different countries' labor markets. It is a comparison between teammates. ENTRA Q2 2026 Remote Work Survey data (n=1,240) documents AI teams at global-remote-first employers where individuals with the same job title, reporting to the same manager, working on the same project, and holding the same performance rating receive cash compensation that differs by 40 to 60 percent. The differential is a function of the country in which the employee's employment contract is registered, not of their role, seniority, contribution, or performance.
The Gulf remote-work architecture intensifies the geographic arbitrage in a specific direction. HUMAIN and G42, the two largest Gulf-headquartered AI employers by headcount growth in H1 2026, are actively hiring remote ML engineers in Eastern Europe and South Asia at all-in package levels of $180,000 to $260,000 (USD, tax-free in the UAE for G42's Abu Dhabi-registered employees and in Saudi Arabia for HUMAIN). The same roles supporting Gulf AI infrastructure, when staffed by San Francisco-based in-office engineers at comparable frontier labs, carry total compensation of $350,000 to $500,000, per ENTRA Q2 2026 Salary Survey data for the frontier-lab tier. The gap is not a function of productivity or output differential. It is a function of labor market supply in the candidate's geography.
The following table illustrates the geographic compensation range for a nominally equivalent Senior ML Engineer role across markets actively represented in ENTRA Q2 2026 Remote Work Survey data.
| Location | Employment structure | Annual cash base (USD equivalent) | Differential vs SF baseline | |---|---|---:|---| | San Francisco (in-office, frontier lab) | Local contract | $210,000-$280,000 | Baseline | | Paris (remote, EU employer of record) | Local contract (EUR) | $118,000-$155,000 | 48% lower at midpoint | | Warsaw (remote, EU employer of record) | Local contract (PLN/EUR) | $130,000-$165,000 | 42% lower at midpoint | | Abu Dhabi (G42, tax-free) | Local contract (AED) | $163,000-$259,000 | 22% lower at midpoint, pre-tax | | Riyadh (HUMAIN, tax-free) | Local contract (SAR) | $180,000-$260,000 | 12% lower at midpoint, pre-tax | | Bangalore (remote, India employer of record) | Local contract (INR) | $85,000-$130,000 | 58% lower at midpoint |
Sources: ENTRA Q2 2026 Remote Work Survey (n=1,240); Hugging Face published compensation philosophy; G42 and HUMAIN offer data per ENTRA Gulf compensation tracking; EUR/USD 1.09, GBP/USD 1.27, SAR/USD 3.75 (canonical August 2026 rates). Tax-free designations are for UAE and Saudi Arabia personal income tax only.
The same-team scenario that the brief survey data produces: a team of eight ML engineers at a global-remote-first AI company, all holding the title "Senior ML Engineer," all reporting to the same lead researcher, all working on the same model infrastructure project. The SF-based engineers at midpoint earn $245,000 in cash base. The Warsaw-based remote engineer, hired two years ago under an EU employer-of-record structure, earns approximately $147,000. The Bangalore-based engineer, joined eighteen months ago, earns approximately $108,000. The spread across that eight-person team is $137,000 in annual cash base, a range larger than the posted band for the role. The posted band does not capture this. The posted band is consistent with every salary paid. That is the design.
The Equity Framing and Its Limits
Geographic pay differentiation is a legitimate and well-established practice. Cost-of-living variation across cities and countries is real. Labor market supply varies by geography. These factors justify geographic differentiation in compensation. The equity issue is not that San Francisco-based engineers earn more than Warsaw-based engineers. The equity issue is that the current disclosure regime provides no instrument for candidates, employees, or external observers to quantify the within-team geographic gap, evaluate whether it is proportionate to cost-of-living and market-supply differences, or identify whether the geographic differential is being applied consistently or in ways that correlate with other protected characteristics.
Buffer, the social media management software company, has published its entire compensation formula publicly since 2013, including the geographic multipliers it applies and the reasoning behind them. Buffer's model applies explicit cost-of-living adjustments using a published formula, makes those adjustments visible to all employees and candidates, and updates them on a defined schedule. No AI employer at the frontier or mid-market tier has replicated this model in full. The closest analogs are Stripe, which publishes level-specific bands without geographic disaggregation, and Hugging Face, which discloses the geographic philosophy and the resulting ranges by country of posting without an explicit formula. Both are materially more transparent than the modal AI employer. Neither resolves the within-team equity question for employees who cannot see their teammates' compensation.
4. What Real Equity Requires Beyond the Band
The evidence from Salary Transparency Month 2026 points toward a disclosure architecture that does not yet exist at any AI employer at scale, but that the regulatory trajectory and the competitive dynamics of the global AI talent market are pushing toward. Four specific elements distinguish genuine equity disclosure from the current band-posting baseline.
Level-Specific Bands
The most immediate structural fix is the one closest to the current compliance architecture: requiring or voluntarily adopting salary bands that correspond to specific career levels rather than career-spanning ranges. Stripe's public compensation structure, which publishes distinct ranges for each of its engineering levels, demonstrates that the operational cost of level-specific disclosure is manageable. Stripe's disclosed senior-IC band for its San Francisco engineering roles runs approximately $30,000 from floor to ceiling. An early-career candidate encountering that band knows precisely where they are likely to be placed. They cannot be anchored to a floor that applies to someone five levels below them.
The EU Pay Transparency Directive transposition, which Germany and the Netherlands are both targeting for early-to-mid 2027, will require level-specific or criteria-specific bands for covered employers. For AI companies with substantive EU operations, this creates a compliance driver for level-specific disclosure that will arrive inside the 2027 planning horizon. The companies that implement it voluntarily in 2026 will not face a structural implementation sprint in 2027. They will also close the career-stage compression effect documented in ENTRA's survey data.
Gender-Disaggregated Pay Data by Role
The California SB 1162 pay data cycle provides a regulatory model for the second element: public or regulator-accessible data showing the distribution of pay by gender within each role category. The current UK gender pay gap reporting requirement produces mean and median gap statistics but not role-level disaggregation. EU Directive 2023/970/EU, when transposed, requires employers with 250 or more employees to report gender pay gaps by "worker category" including pay grade and pay band. For AI employers, "worker category" reporting at the pay grade level would reveal within-band gender concentration — the core finding obscured by the current posting-only regime.
Hugging Face (85/100 ENTRA Pay Transparency Audit Q2 2026), Stripe (82/100), and GitHub (75/100) represent the current leaders on overall transparency score across ENTRA's 87-company audit cohort. All three score strongly on Salary Band Disclosure (30 points) and Compensation Philosophy Public (15 points). None currently publishes gender-disaggregated pay data by role category on a voluntary basis, which explains why the maximum score in the cohort remains below 90: the Internal Pay Equity Data Published dimension (20 points) is the most commonly deficient across all 87 companies screened.
Geographic Equity Benchmarks
The third element is the most novel and the least developed in either regulatory or voluntary frameworks: a mechanism for publishing the geographic adjustment rationale and its application, so that employees and candidates can evaluate whether the differential between their location and the company's benchmark market is proportionate to cost-of-living and market-supply differences, and whether it is applied consistently.
Buffer's openly published formula represents one model. An employer-agnostic benchmark model would use an external cost-of-living index (such as the Economist Intelligence Unit's cost-of-living data or Numbeo's composite index) with a defined application methodology. The result would not be a single global compensation rate — it would be a transparent formula that produces the geographic differential rather than a differential that arrives at the offer stage without visible derivation. ENTRA's Q2 2026 Remote Work Survey found that 67 percent of remote AI practitioners working for global employers did not know the geographic adjustment formula their employer used to set their compensation, and 71 percent did not know whether the same formula applied to their teammates in other locations.
The 2027 Milestone and Who Is Closest
The EU Directive transposition cycle creates a hard deadline for two of these four elements across all 250-plus-employee EU employers: level-specific (or criteria-specific) bands and gender pay gap reporting by worker category. For Hugging Face (approximately 420 EU-based employees), Mistral (approximately 380), and any US frontier lab with a substantive EU research hub, the first EU-compliant disclosures will need to be operational before the first enforcement actions, which member state equality bodies are expected to initiate in H1 2027.
The gap between the current leader (Hugging Face, 85/100) and a hypothetical full-equity-disclosure score of 100 is concentrated in two dimensions: Internal Pay Equity Data Published (ENTRA scores Hugging Face at 14/20 on this dimension, reflecting the absence of role-level gender pay gap public data) and Market-Rate Commitment / External Benchmarking (11/10 — maximum score — for publishing compensation philosophy with explicit market anchoring, but no publicly accessible formula for geographic adjustment). The score ceiling, absent regulatory compulsion, is around 88 to 90/100 for the current leading cohort. The regulatory floor that EU Directive enforcement will create in 2027 will require companies to close the gap to approximately 80/100 on ENTRA's scoring framework regardless of voluntary posture.
5. The Forecast: Three Scenarios for 2027
The three hidden dimensions documented in this report — gender composition, career-stage compression, and geographic arbitrage — will each face different pressures in 2027. The outcome depends on which of three plausible scenarios governs the next eighteen months of regulatory and market development.
Scenario A: Regulatory Escalation Closes the Quality Gap
Under Scenario A, EU Directive enforcement arrives on the aggressive end of the projected timeline, with member state equality bodies bringing enforcement actions against systematic non-compliers in H1 2027. This triggers a wave of level-specific band adoption and gender pay gap reporting by role across the 250-plus-employee EU employer tier. California SB 1162's second annual filing cycle (CY2026 data, due May 2027) shows a measurable narrowing of the within-band gender gap at tech employers responding to the June 2026 CRD aggregate data, with the unadjusted tech-sector gap declining from 17.4 percent toward 14 to 15 percent — a trajectory consistent with Arjuna Capital's research on companies subject to active shareholder engagement on pay equity, which showed an average within-band gender gap reduction of 6 to 8 percentage points per year among engaged companies. Geographic equity benchmarks begin appearing in company compensation pages as a competitive recruitment differentiator in the EU market, where candidates compare geographic multiplier methodologies across employers.
The within-band gender composition gap begins closing at scale. Career-stage compression becomes legally non-compliant for EU-operating employers. The geographic arbitrage narrows in the EU context but persists in the US and Gulf markets where no equivalent regulatory pressure exists.
Scenario B: AI Talent Concentration Widens Every Gap
Under Scenario B, the acceleration of foundation model capability investment through 2026 and 2027 concentrates AI talent further at the frontier-lab tier, widening the compensation gap between the top-of-band roles (where demand is sharpest and candidates are fewest) and the floor-of-band roles (where supply is larger and compensation increases are modest). Band widths grow rather than narrow, because frontier labs respond to multi-level talent concentration by posting compliance-minimum ranges that contain an ever-wider span of career stages within a single disclosed number. The within-band gender gap widens because the fastest-growing compensation element is equity and above-base performance pay in the role categories where women are most underrepresented (Foundation Model Research, Inference Optimization). SB 1162's second filing cycle shows a widening unadjusted gap as equity-driven total compensation pulls the male mean further above the base pay captured in the mandatory reporting.
Companies achieve 100 percent band-posting compliance across all mandated jurisdictions while the equity distribution inside those bands deteriorates. The transparency regime records continued improvement in posting rates while the structural inequity it was designed to address accelerates.
Scenario C: Remote Work Normalization Closes the Geographic Gap but Stalls Everything Else
Under Scenario C, the competitive pressure documented in ENTRA's Q2 2026 Remote Work Survey — where 67 percent of remote practitioners do not know their employer's geographic adjustment formula — drives a voluntary shift toward global-benchmark compensation at a material minority of AI employers. Following Buffer's model, four to six mid-market AI companies (ENTRA estimates Hugging Face and Stripe as the most likely early movers based on existing compensation philosophy posture) publish explicit geographic equity benchmarks: a defined cost-of-living index, a published multiplier table, and a commitment to narrow the geographic differential to a maximum percentage of the benchmark market rate. Geographic pay gaps begin closing within those employers' distributed teams. The broader market watches but does not follow without a regulatory requirement.
The gender composition and career-stage compression problems stall under Scenario C. Without the regulatory driver that EU Directive enforcement creates, voluntary disclosure of within-band gender pay distribution remains optional and commercially sensitive. Level-specific bands spread slowly among employers already competing on transparency as a recruitment brand (Stripe, GitHub, and their peers) but do not reach frontier labs or the Gulf AI employer tier without a specific compliance event.
The most probable 2027 outcome is a combination of all three scenarios, operating across different employer tiers simultaneously: regulatory escalation closing the gap at EU-operating 250-plus-employee employers, talent concentration widening the gap at US-only frontier labs operating under compliance-minimum disclosure, and geographic equity normalization beginning at a small leading cohort of global-remote-first employers. The net result is a more fragmented equity landscape, not a converging one, despite continued improvement in global band-posting rates.
The measure that would move all three dimensions simultaneously does not yet exist in any jurisdiction: a disclosure requirement that mandates the publication of actual pay distribution by gender and career stage within each disclosed salary band, alongside the geographic adjustment formula applied to remote and internationally contracted employees. California SB 1162 achieves the gender-distribution element at the aggregate sector level. The EU Directive achieves the level-specificity element for EU-operating employers. No framework yet combines them with geographic equity disclosure. The combination is the next frontier of pay transparency legislation, and the Salary Transparency Month 2026 data makes a strong empirical case that it is needed.
Methodology
ENTRA Pay Transparency Audit Q2 2026: 87 AI companies screened across five dimensions: Salary Band Disclosure (30 points), Equity Compensation Transparency (25 points), Internal Pay Equity Data Published (20 points), Compensation Philosophy Public (15 points), and Market-Rate Commitment / External Benchmarking (10 points). Scoring windows: January 1 to June 30, 2026 for pay equity reports, DEI filings, and annual reports; May 1 to July 20, 2026 for job posting and careers-page audit. Twenty companies scored for the August 2026 ENTRA Pay Transparency Index; this report draws on the broader 87-company screening cohort for non-indexed analysis. Gender composition analysis in Section 1 uses role-category employment composition data collected during the same audit window through publicly available diversity and pay equity reports, corroborated against California CRD pay data filings where available. Scores cited for Hugging Face (85/100), Stripe (82/100), and GitHub (75/100) are ENTRA Pay Transparency Audit Q2 2026 final scores.
ENTRA Q2 2026 Salary Survey: 2,140 AI practitioners surveyed globally between April 5 and June 30, 2026. Respondent composition: 61 percent US-based, 18 percent European, 12 percent Asia-Pacific, 9 percent Gulf/MENA. Career-stage methodology: respondents self-reported years of professional AI experience post-degree and were classified into early-career (fewer than 3 years), mid-career (3 to 8 years), and senior (more than 8 years) cohorts. Compensation delta for Section 2 computed as the cohort mean final offer total compensation for early-career versus mid-career respondents within matched role titles and employer tiers. The 34 percent gap and the $78,000 to $95,000 absolute delta represent averages across the Senior ML Engineer and Research Scientist title categories; by employer tier, the applied-AI tier shows the $78,000 delta and the frontier-lab tier shows the $95,000 delta. Opening offer behavior analysis uses the subset of respondents (n=1,247) who reported both an opening counter-offer amount and a final offer amount for their most recent completed hiring process.
ENTRA Q2 2026 Remote Work Survey: 1,240 AI practitioners employed by global remote-first or remote-eligible AI employers, surveyed between April 12 and June 28, 2026. Geographic cash compensation gap methodology: respondents reported annual base cash compensation in local currency, converted to USD at canonical exchange rates (EUR/USD 1.09, GBP/USD 1.27, SAR/USD 3.75, all prevailing Q2 2026 rates per respective central bank data). Gap computed as the percentage difference between each respondent's USD-equivalent base and the median USD base reported by same-employer, same-title respondents located in San Francisco. The 40 to 60 percent range represents the interquartile range of the geographic cash gap across the full cross-geography cohort. The 48 percent specific figure for the Hugging Face Paris-to-SF comparison uses ENTRA Q2 2026 Remote Work Survey actual offer distributions (n=31 Hugging Face respondents) rather than posted band midpoints. The 67 percent and 71 percent figures on awareness of geographic adjustment formulas are from the same survey. All respondents consented to anonymized aggregated reporting.
External sources: California Civil Rights Department 2025 Pay Data Report (aggregate professional computing occupations data, published June 2026); McKinsey and LeanIn.org "Women in the Workplace 2025" (September 2025, senior IC representation in technology sector); American Association of University Women 2026 STEM Pay Analysis (computing and mathematical occupations gender earnings gap); Amazon UK Gender Pay Report 2025; Meta UK Gender Pay Report 2025; Microsoft UK Gender Pay Report 2025; Levels.fyi Q2 2026 composite for PhD-direct AI research offers at frontier labs; Hugging Face public careers page compensation disclosures (Paris and San Francisco bands, July 2026); Buffer public compensation formula (buffer.com/salaries, current as of Q2 2026); Arjuna Capital research on pay equity outcomes at shareholder-engaged companies (Arjuna Capital Pay Equity Engagement Annual Update, 2025); EU Pay Transparency Directive 2023/970/EU; Germany Entgelttransparenzgesetz (EntgTranspG), enacted July 2017, effective January 2018 for individual pay information rights; Stripe public engineering compensation disclosures (stripe.com/jobs, Q2 2026 audit).
The ENTRA Q2 2026 Salary Survey is conducted independently by ENTRA Intelligence. No compensation is accepted from companies included in the ENTRA Pay Transparency Audit. Company transparency scores cited in this report are editorial assessments based on publicly available data. All company-level gender pay gap figures are reproduced from companies' own mandatory filings under UK and California statutory reporting requirements and are attributed accordingly. ENTRA Intelligence is not responsible for the accuracy of company-filed statutory data. Exchange rates: EUR/USD 1.09, GBP/USD 1.27, SAR/USD 3.75 (canonical August 2026 rates).
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.