Colorado's Equal Pay for Equal Work Act logged 2,800-plus complaints and $841,500 in fines through June 2026, pushing frontier AI labs to post salary ranges on every role visible to state applicants. Anthropic lists $120,000 to $320,000 in base pay for senior machine learning engineers. OpenAI posts $100,000 to $300,000-plus for comparable titles. That $200,000 spread satisfies the letter of the law and tells a candidate almost nothing about where their offer will land.
The Law Got What It Asked For
Colorado's Part 2 EPEWA amendments, effective January 1, 2024, extended salary disclosure requirements to any role "reasonably expected to be performed in Colorado," which in practice captured every remote-eligible posting by any US employer. The Colorado Department of Labor and Employment has enforced it with measurable results. Through June 2026, CDLE reported 2,800-plus Part 2 complaints filed, 600-plus voluntary compliance letters issued, 24 formal citations, and total fines of $841,500 across all enforcement actions. Individual violation penalties run $500 to $10,000 per non-compliant posting. New York City's Local Law 32 followed a parallel track: 2026 amendments tightened the "good-faith range" standard and explicitly barred "$1 to $1 million" placeholder postings that had been common through early 2025.
Frontier labs complied. Anthropic, OpenAI, and xAI now post salary ranges on every public-facing US role. The compliance is real. The transparency is not.
Anthropic's posted range for a Senior Research Scientist in San Francisco runs $120,000 to $320,000 in annual base pay. OpenAI lists base salary from $100,000 to $300,000-plus, described in postings as varying "depending on skills and experience." xAI research roles visible to Colorado applicants in Q2 2026 ran $130,000 to $310,000 in base. Those three numbers share a structural feature: each spans roughly $180,000 to $200,000.
The ENTRA Job Signal Index Q2 2026 found an average posted band width of $180,000 to $220,000 across senior ML and research roles at frontier labs. That average is not accidental. Labs set bands wide enough to cover multiple seniority sub-levels collapsed under one job title (staff, senior staff, and principal all filed as "Senior Research Engineer"), to accommodate both San Francisco placements and remote candidates in lower-cost markets under a single posting, and to preserve enough flexibility to avoid equal pay exposure if two people with the same title land at different points in the band. The compliance standard does not require a lab to disclose the median landing point for a candidate with your profile. It requires a band. The labs provide one.
Where the Band Ends and the Negotiation Begins
The posted band functions as an anchoring instrument even when it obscures distribution. ENTRA Q2 2026 survey data, drawn from 214 completed US-based AI and ML role placements between January and June 2026, found that candidates who entered compensation discussions with a specific band anchor received starting offers 12 to 18 percent above candidates who negotiated without one. At a $600,000 total-compensation package, that gap runs $72,000 to $108,000 per year in realized value.
The distance between a posted base salary range and realized total compensation is where most candidates lose money. A compensation consultant who advises senior AI researchers across three Bay Area labs, speaking on background, described the architecture: "The base range on the posting is real. But it's the floor of their flexibility, not the ceiling. Equity is where the negotiation actually lives, and equity doesn't appear in any posting anywhere."
That framing is precise. Levels.fyi self-report data aggregated through Q2 2026 places median total compensation for Anthropic L5 engineers near $625,000 against a posted base ceiling of $320,000. At OpenAI, L5 total compensation sits near $1.15 million per Levels.fyi, against a posted base ceiling of $300,000-plus. The delta is not fraud; it is architecture. Equity, refresh grants, signing bonuses, and milestone payments all sit outside the salary range that CO EPEWA and NYC LL32 require labs to disclose. California's annual pay data reporting requirement adds a second data surface — employers must publish median and mean pay by job category, disaggregated by gender and race — but that disclosure is backward-looking and aggregate, not a predictor of any individual offer.
European AI labs operating in comparable technical markets take a structurally different approach. Mistral, hiring in Paris, posts ranges in the €108,000 to €142,000 band for comparable seniority levels — a €34,000 spread against the $200,000 equivalent at US frontier labs. Hugging Face Paris engineering roles rarely carry ranges wider than €40,000 (ENTRA Job Signal Index monitoring, EU AI employer postings, Q1–Q2 2026). The EU Pay Transparency Directive, which took effect June 7, 2026, codified what was already standard practice: European AI employers treat narrow bands as a credibility signal. "A band of €50,000 to €120,000 tells candidates you haven't done the work," one Paris-based technical recruiter told a 2025 TechHR Europe panel. The legal architecture across the Atlantic is converging. The behavioral norm in US frontier hiring is not yet following.
Three Numbers to Watch
CDLE's per-citation ceiling. At $10,000 per violation, a lab posting 500 roles with a non-compliant range faces $5 million in theoretical exposure. CDLE's largest single enforcement action through June 2026 remained in the low six figures. The agency has not yet issued a precedent-setting large fine against a frontier tech employer. If CDLE closes that gap before year-end — and the complaint volume suggests it has the material to try — the calculus for employers posting wide bands at scale changes.
NYC's undefined band-width threshold. The 2026 amendments to Local Law 32 closed the placeholder loophole but did not define what constitutes a good-faith range. The NYC Commission on Human Rights stated in April 2026 public guidance that it intends to scrutinize ranges that do not meaningfully constrain employer discretion — language that stops short of setting a numerical limit. If the Commission issues formal guidance establishing a maximum band width as a percentage of midpoint (a standard used in some EU jurisdictions), frontier labs posting $200,000 spreads on roles with a $250,000 midpoint would face direct compliance exposure without changing their current offer behavior at all.
Voluntary band compression as a recruiting signal. No frontier lab has committed publicly to narrowing its posted ranges. The competitive incentive runs the other way: wider bands preserve maximum flexibility at offer time. But a lab that posts $200,000 to $260,000 instead of $100,000 to $300,000-plus signals to a senior researcher that the offer will land in a predictable range. At the talent tier where Anthropic and OpenAI compete, that predictability may be worth more than the flexibility the wide band buys. Watch whether any lab tests compressed bands in Q4 2026 role postings, particularly for hard-to-fill interpretability and alignment research roles where candidate leverage is highest.
A $200,000 spread is not non-compliance; it is the legal structure doing exactly what it was designed to do, and knowing that is the only edge a candidate has going into the room.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.