Top 20 — AI Employers · Pay Equity Progress

Posting a salary band is not the same as closing a pay gap

Across 20 prominent AI employers, only five reach the AA threshold for pay equity progress. The AI industry has made genuine progress on disclosure: more companies are posting pay bands, more jurisdictions are mandating it, and employees have more information than they did two years ago. But transparency and equity are not the same thing. Knowing the band you were hired into tells you nothing about whether the person next to you, doing the same job at the same level, is being paid the same amount. This ranking measures the harder question: which AI companies are actually closing internal compensation gaps, conducting audits with teeth, correcting inequities when they find them, and publishing enough information to be held accountable?

Only five companies in this ranking score at the AA threshold or above, meaning they have demonstrated a combination of rigorous internal auditing, measurable gap closure, corrective action programs backed by real dollar commitments, and public-facing data that would satisfy a skeptical institutional investor or an informed job candidate. The remaining 15 range from strong processes with incomplete disclosure (A tier) to compliance-minimum postures with no voluntary disclosure (B tier). The divergence is not primarily a function of company size or resources. It is a function of institutional will.

What transparency measures, and what it misses

The August 1 ENTRA Pay Transparency Score ranking assessed a related but distinct question: do AI companies disclose salary bands publicly? That dimension matters enormously for candidates deciding where to apply. But external disclosure does not tell you whether the bands are applied equitably across protected classes, whether manager discretion has been constrained, whether an audit has ever been run, or whether anyone has written a corrective check when gaps were found.

Internal pay equity is the discipline of ensuring that employees doing the same work at the same level are compensated at the same rate regardless of gender, race, or other protected characteristics. It requires a different methodology than salary posting: regression analysis controlling for role, level, geography, and tenure; testing for statistically significant residual gaps after controls; funding corrections when gaps are found; and doing this on a regular cycle rather than as a one-time exercise. It also requires publishing enough of the methodology and results that the work can be evaluated, not just asserted.

The AI industry, which has grown faster than almost any sector in history and is still largely exempt from the regulatory pressure that applies to larger and more mature employers, has not made this a priority at scale. The scoring in this ranking reflects that gap directly.

What the AA tier does differently

The five companies that reach the AA threshold -- Microsoft, Salesforce, Stripe, GitHub, and Google -- share three structural characteristics that distinguish them from the rest of the cohort.

First, they audit on a regular cycle with a defined methodology. Microsoft's annual pay equity analysis covers 190 countries, applies a consistent regression model, and publishes results. Salesforce has done the same every year since 2016. Google's analysis covers all Alphabet entities including DeepMind. These are not one-time exercises conducted under crisis pressure. They are institutional programs with named owners, budgets, and publication commitments.

Second, they fund corrections. Salesforce has paid $22M in cumulative pay equity corrections since 2015 -- the highest publicly disclosed corrective figure in this cohort and one of the highest in the technology sector. Microsoft funds in-cycle adjustments, meaning corrections happen at the point of discovery rather than being deferred to the next annual review cycle. Stripe's 2025 ESOP audit produced targeted cash adjustments for 340 employees. These numbers are small relative to total payroll, but the practice of making corrections at all, and disclosing that corrections were made, is what separates the AA tier from everyone below it.

Third, they structurally constrain manager discretion. Stripe's compensation framework now ties manager performance reviews to equity outcome metrics: a manager whose direct reports show statistically significant pay dispersion by protected class faces a documented review. Microsoft's offer-generation process applies band constraints that limit the room for individual manager judgment to compound into systemic gaps. Google's offer calibration panels include equity review checkpoints. The common thread is that equity is not left to individual goodwill; it is built into the process architecture.

GitHub earns its AA score primarily through its position inside Microsoft's equity infrastructure, augmented by subsidiary-level disclosure at the product-team level. Copilot team equity data is disclosed separately from the broader GitHub entity -- a granularity of reporting that is unusual and valuable. It is the most derivative score in the top five, but derivativeness from a strong program is still a strong score.

The audit-to-action gap

The BBB tier -- Meta AI, Anthropic, and OpenAI -- contains companies with the organizational capacity to run rigorous equity programs but where the gap between audit and action is either not closed or not disclosed.

Meta's equity methodology is among the most technically sophisticated of any company in this cohort. Its historical adjusted gap data shows genuine improvement year over year. The problem for this scoring window is that the 2026 annual equity report was delayed from its normal Q1 publication into Q2-Q3 2026 as a consequence of the 2024-2025 DEI restructuring and the resolution of DOJ-related inquiries. A delayed report is not the same as a bad program, but it is a disclosure failure. ENTRA will rescore Meta when the 2026 report publishes.

Anthropic and OpenAI are both private companies whose equity disclosure obligations are defined primarily by California SB 1162 -- which requires pay data filing but not an equity audit. Both companies pay at the top of the market, which provides some structural protection against the most egregious forms of pay inequity. But paying well and paying equitably are not the same thing. A company can have median total compensation above $400K and still have significant within-level disparities by gender. Without an audit, neither the company nor its employees can know. The anticipated IPO paths for both companies will force this issue: institutional ESG investors will require audits as a condition of participation, and public benefit corporation governance for OpenAI introduces new fiduciary considerations around workforce equity.

Scale-up equity debt: The BB and B tier

Below the BBB line, the ranking includes eight companies in the BB range and five in the B range. The common thread is not that these companies are acting in bad faith. It is that they are scaling faster than their equity infrastructure.

Databricks, at a $10B+ Series J valuation, files California SB 1162 data and conducts internal reviews. Cohere operates across three jurisdictions with different obligations and has not centralized its reporting. Scale AI's workforce structure -- with a significant contractor population alongside FTE engineers -- creates genuine methodological complexity for equity analysis: the contractor layer is not subject to the same disclosure obligations as the FTE population, and combining the two into a single equity figure would require methodology that Scale AI has not published. Mistral is transitioning from a sub-50-employee French company (below the mandatory reporting threshold) to a 200+ employee organization that will trigger full Index d'Egalite Professionnelle obligations for its 2026 reporting year.

The five B-tier companies -- Weights and Biases, ElevenLabs, Replicate, Together AI, and Wayve -- are at stages of scale where formal equity infrastructure is only just becoming relevant. Replicate and Together AI have under 100 FTEs. ElevenLabs grew from 50 to 200+ employees in 18 months. Wayve is preparing for its first mandatory UK Gender Pay Gap filing. The B rating is not a judgment that these companies are inequitable. It is a reflection that their equity programs, such as they exist, are not yet at a stage where they can be evaluated from outside the company.

What comes next: IPO pressure, EU enforcement, and convergence

Three structural forces are converging to make pay equity progress a requirement rather than a choice for AI companies over the next 24 months.

The first is IPO pressure. Anthropic, OpenAI, Databricks, Stripe, and Cohere are all plausible IPO candidates in the 2026-2028 window. Institutional investors -- particularly large sovereign wealth funds, pension funds, and ESG-mandate index funds -- now routinely require equity audit disclosure as part of due diligence. The S-1 process will surface these questions in a public context. Companies that have not built equity programs by the time they file will be scrambling to produce reports under conditions that are far less controlled than voluntary pre-IPO programs.

The second is EU enforcement. The EU Pay Transparency Directive, which entered into force in June 2023 with a transposition deadline of June 2026, requires employers with 250 or more employees in EU member states to publish gender pay gap data annually and to conduct joint pay assessments when gaps exceed 5%. For AI companies with EU operations -- Google DeepMind in London and Zurich, Mistral in Paris, Wayve in London, Cohere in London -- enforcement is no longer a future prospect. It is a present obligation.

The third is the convergence of transparency and equity as a single market-facing signal. Salary band disclosure has become a norm. The next competitive differentiator in the AI talent market will be companies that can credibly say: not only do we disclose what we pay, but we have verified that we pay everyone doing equivalent work at equivalent rates. That is a statement only an AA-tier company can currently make. The others have, at most, 18 months to catch up before the gap becomes a material talent disadvantage.

How we ranked

The Top 20 AI Companies by Pay Equity Progress 2026 is scored across 5 dimensions, equally weighted at 20 points each:

• Internal Pay Equity Audit Completion (20 pts): whether a third-party or internally audited pay equity program exists, how frequently it runs, and whether results are published. Source: company-published equity and ESG reports; EEOC EEO-1 filings; California SB 1162 and Washington SB 5761 compliance records; EU Pay Transparency Directive filings (Q2 2026).

• Gender Pay Gap Closure Rate (20 pts): year-over-year improvement in adjusted and unadjusted gender pay gaps. Source: UK Gender Pay Gap Service mandatory filings; Glassdoor pay gap ratings; company-published adjusted gap figures from annual equality or ESG reports (trailing 24 months).

• Pay Band Parity Across Levels (20 pts): evidence that IC and manager pay bands are applied without discrimination by protected class; absence of unlisted discretionary bands. Source: Levels.fyi public salary distribution analysis; ENTRA Salary Survey H1 2026 (n=2,400+ self-reported); voluntary internal band documentation.

• Equity Corrective Action Implementation (20 pts): documented salary adjustments made as a result of equity audits; dollar value of corrections disclosed; frequency of adjustment cycles. Source: named CHRO statements; SEC or proxy disclosures; Glassdoor verified employee accounts.

• Pay Equity Data Transparency (20 pts): public-facing pay equity or equality report with disclosed methodology; adjusted and unadjusted gaps both reported; intersectional data available. Source: ENTRA Pay Equity Monitor Q2 2026 (n=47 AI employers screened).

Data window: January 2024 to June 2026 (primary window Q1 2025 to Q2 2026); corrective action history reviewed trailing 36 months.

Sample size: 47 AI employers longlisted; 20 scored on the full 5-dimension model; 2,400+ salary data points from ENTRA Salary Survey H1 2026; 20 company equity or ESG reports reviewed; 14 regulatory filings cross-referenced.

Rating bands: 90-100 AAA | 80-89 AA | 70-79 A | 60-69 BBB | 50-59 BB | below 50 B.

Limitations:

• Private companies (Anthropic, OpenAI, Databricks, Cohere, Mistral, Scale AI, Replicate, Together AI, ElevenLabs) have no obligation to publish equity audit results outside jurisdiction-specific minimums. Their scores reflect available proxy signals and are structurally capped relative to public-company peers with mandatory disclosure obligations.
• Adjusted pay gap figures reported by companies use company-defined control variables. Cross-company adjusted gap comparisons are directional, not precise -- methodology differences may over- or under-state true gap closure relative to a consistent external standard.

Inquiries about methodology: methodology@entracareers.com