Three US states require posted salary ranges on AI job listings; August 2026 is the compliance deadline for mid-size AI companies filing their first California pay data reports with the DFEH. The numbers landing in public record show L5 machine learning engineer bands spanning $140,000 in spread at Anthropic and $200,000 at OpenAI. What engineers can learn from those ranges, and what the ranges are designed to obscure, are different questions.
The Three-State Framework
Colorado's Equal Pay for Equal Work Act (EPEWA) applies to every employer with at least one Colorado-based employee and covers any job posting accessible to Colorado applicants. That language has reach: a tech company that posts an engineering role nationally, even on its own careers page, must include a salary range and benefits summary if any Colorado resident might see it. The scope effectively requires frontier labs to disclose ranges on every national posting with no carve-out mechanism. The Colorado Department of Labor and Employment reported as of June 2026 that it had received more than 2,800 complaints under Part 2 of the Act, issued more than 600 voluntary compliance letters, and collected fines totaling more than $800,000. Individual penalties run $500 to $10,000 per violation. Effective July 1, 2026, CDLE's jurisdiction to investigate wage complaints increased from $7,500 to $13,000 per employee, a threshold that rises every two years. The enforcement apparatus is not static, and the stakes for a lab running 40-plus open ML positions in a single quarter compound quickly.
New York City's Local Law 32, in force since November 2022, targets a different pressure point. Any private employer with four or more employees must include a minimum and maximum salary in every job advertisement. The provision that matters most for AI companies is the remote-work clause: a role that would report to a New York City supervisor or office triggers the disclosure requirement regardless of where the hire physically works. For frontier labs that offer remote-eligible positions to maximize candidate pools, and that maintain New York presences, virtually every national engineering posting falls within scope. Anthropic lists a New York office. OpenAI's East Coast footprint has grown alongside its planned expansion to 8,000 employees by December 2026. The legal reading is direct: post the range, or document a defensible New York exemption for each role.
California's SB 1162 operates on two distinct tracks. Any employer with 15 or more employees must include a pay scale in every job posting. Any employer with 100 or more employees must submit annual median pay data to the DFEH covering mean and median hourly wages by job category, race, ethnicity, and sex. For the 2026 reporting cycle, California expanded the required categories from 10 to 23, adding classification groups that map more precisely to the roles AI labs fill: machine learning research, AI safety, model evaluation, and infrastructure engineering now appear as distinct reporting categories rather than being collapsed into a generic "Computer and Mathematical" bucket. Companies that first crossed the 100-employee threshold in 2025 and have not previously filed are finalizing inaugural submissions, with DFEH accepting first-time reports through August 2026. First-failure penalties run $100 per employee. A 200-person AI company facing its first enforcement action carries $20,000 in exposure, a number that establishes the penalty baseline for every subsequent filing year.
What AI Lab Disclosures Reveal
The table below draws on job postings tracked in the ENTRA Salary Survey Q2 2026 and cross-referenced against Levels.fyi self-reported compensation data. All figures represent total compensation (TC), meaning base salary plus equity annualized at current company valuation and, where applicable, a target cash bonus. Posted ranges reflect disclosures in public job listings through July 2026, or ENTRA estimates where companies post ranges only in state-compliant jurisdictions.
| Role | Anthropic | OpenAI | Google DeepMind | xAI | |---|---|---|---|---| | Senior ML Engineer (L5) | $280K-$420K | $300K-$500K | $280K-$450K | $320K-$520K | | Mid-Level ML Engineer (L4) | $200K-$340K | $240K-$400K | $210K-$360K | $240K-$380K | | Research Scientist (IC4/L4) | $300K-$480K | $320K-$560K | $290K-$470K | $300K-$490K |
Source: ENTRA Salary Survey Q2 2026; Levels.fyi self-reported data through June 2026. Figures are estimates for total compensation and should be treated as approximate.
Three data points from this table that engineers should act on before applying.
The floors are legally enforceable in California. A company that posts $280,000 as its L5 ML floor in a California-accessible listing cannot offer a California-based hire below that number without creating SB 1162 exposure. The floor is not an aspiration or a starting anchor for negotiation. It is the minimum the employer has publicly committed to for that role in that jurisdiction.
Levels.fyi self-reported data for Anthropic's L5-equivalent software engineering roles shows a median total compensation of approximately $625,000 in recent reporting windows, well above the $420,000 ceiling in Anthropic's posted band. The gap exists because posted ranges reflect base salary plus standard equity grants. Anthropic's equity component is priced against a company valued in private markets at several hundred billion dollars following its 2026 funding rounds; the realized total compensation fluctuates with valuation and cannot be accurately represented in a static job posting. The disclosed range is real. It is also the floor of the actual story, not the ceiling.
xAI's posted band at L5 sits at $320,000 to $520,000, a $200,000 spread that is nominally comparable in width to OpenAI's $300,000 to $500,000 range but operates against a different internal structure. xAI runs fewer formal levels than Anthropic or Google, which compresses the differentiation within any given level. Candidate-reported offer data tracked by ENTRA suggests the actual distribution of xAI L5 offers clusters tighter around the midpoint than Anthropic's or OpenAI's, where offers routinely land near the floor or near the ceiling depending on factors the posting does not disclose.
The Obfuscation Question
A 50 to 70 percent spread between the floor and ceiling of a posted salary band is not a rounding error. On Anthropic's $280,000 to $420,000 L5 range, the ceiling is 50 percent above the floor. On OpenAI's $300,000 to $500,000 band, it is 67 percent. These spreads serve companies in ways that have nothing to do with transparency.
The first function is candidate capture. A company that narrows its posted range to the realistic distribution of offers it actually makes, perhaps $340,000 to $380,000 for a typical L5 ML hire, causes candidates expecting $420,000 to self-select out before the first screen. Wide bands preserve candidate flow across expectation ranges without the recruiter having to reveal what a strong candidate will actually be offered. Compliance is achieved. Signal value is compressed to the point of near-uselessness.
The second function is internal differentiation management. Anthropic's L5 band, per accounts from candidates who have navigated the process and shared details with ENTRA, appears to span two distinct internal sub-ranges: a lower-L5 and an upper-L5, each with different base floors and equity grant parameters. The single public range of $280,000 to $420,000 collapses that distinction. A candidate who receives $295,000 TC and a colleague who receives $405,000 TC can hold identical external job titles, identical levels, and sit within the same public band. The law required the company to show the range. It did not require the company to show the distribution.
Colorado's CDLE has addressed this directly in its enforcement posture. The statute requires that a posted range represent what the employer "actually intends to pay" for the role, and that open-ended or inflated ranges that do not reflect a realistic offer distribution do not satisfy the good-faith standard under EPEWA. Several 2026 CDLE settlement letters cited band width as a compliance concern. The direction of enforcement is clear: the agency is moving from presence-or-absence scrutiny toward distribution-quality scrutiny. A Brightmine survey from late 2025 found that 45 percent of US employers now operate in jurisdictions with active pay transparency laws, and more than 68 percent of job postings included salary ranges, up from 45 percent in 2023. Compliance rates improved. Disclosure quality did not keep pace.
A June 2026 analysis tracking 42,000 live tech job postings found that only 14 percent of listings globally disclosed a salary range, despite the three-state requirements and a growing patchwork of municipal and state rules. In the postings that did include ranges, the average floor-to-ceiling spread for ML engineering roles ran 58 percent. The number tells you the compliance posture of the industry: companies are disclosing the minimum the law requires, at the maximum band width the law permits.
The legislative trajectory through 2027 runs in one direction. The Salary Transparency Act, reintroduced in the US Senate in February 2026, would extend pay range disclosure to all federal postings and establish a private-sector framework overlapping substantially with what Colorado and California already require. Illinois, Washington, Massachusetts, and New Jersey advanced salary disclosure legislation in 2025 and early 2026. A SHRM projection from Q1 2026 estimated that by January 2027 more than 65 percent of the US workforce will operate under some form of pay transparency requirement, up from roughly 45 percent today. For frontier AI labs posting roles into an increasingly covered map, the question is no longer whether to disclose salary ranges. It is how long a 67 percent band spread will satisfy a regulator who has read the EPEWA enforcement manual.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.
Sources: Colorado EPEWA requirements, SixFifty — CDLE EPEWA enforcement 2026, Faegre Drinker — NYC Local Law 32, SHRM — California SB 1162 pay data reporting 2026, Stoel Rives — Anthropic salaries, Levels.fyi — OpenAI software engineer salary, Levels.fyi — xAI software engineer salary, Levels.fyi — Pay transparency trends 2026, Candoriq — Tech job postings salary disclosure analysis, Dev Journal