The same senior ML engineer role, posted on the same day, carries a visible salary band if the hiring company has offices in Colorado or California, and carries nothing at all if that role is posted for Paris, Tel Aviv, or Abu Dhabi. That gap is not a flaw in the regulatory system. That gap is the system. ENTRA's Job Signal Index Q2 2026 tracked more than 12,000 active AI roles at funded startups globally and found that 62% post no salary range anywhere: not in US jurisdictions with active enforcement, not on LinkedIn UK, not in the EU member states with the Directive's transposition deadline already passed. Salary Transparency Month, now in its second week, is mapping the arbitrage that gap creates, country by country.
This Week's Signal
Five data points published across ENTRA's bureaus last week anchor the pattern for Week 2.
The $43K uplift. Friday's report, "Pay Transparency Dividend: AI Engineers Who See Bands Earn $43K More," quantified the candidate-side cost of opacity. ENTRA Q2 2026 survey data (n=214 US-based AI and ML placements, January through June) found a median $43,000 gap in realized first-year total compensation between candidates who entered offer negotiations with access to verified salary band data and those who did not. The gap is not a negotiation-skill gap. Candidates who cannot see the band cannot anchor to it.
The 37% narrow-band rate. Saturday's Top 20 AI Roles by Salary Band Transparency Quality index rated every covered AI title across 12 jurisdictions. Thirty-seven percent of roles globally post what ENTRA classifies as a narrow, verifiable band: a floor-to-ceiling spread below 40% of midpoint, in a jurisdiction with active enforcement. The remaining 63% post either no range, a placeholder spanning more than $200,000, or a range visible only in one compliant jurisdiction with no disclosure elsewhere.
The Anthropic case. Sunday's company deep-dive found Anthropic's posted band ceiling at $320,000 for senior base salary, against ENTRA recruiter network intelligence placing L5 total compensation at approximately $625,000. The $305,000 gap reflects a disclosure architecture that covers base salary and omits equity, refresher cadence, and performance-linked grant structure. Anthropic scores AAA (98/100) on the ENTRA Talent Index and BBB (62/100) on the ENTRA Pay Transparency Index. Both ratings are accurate. One measures employer quality. The other measures whether candidates can see clearly before they negotiate.
NYC at 26% non-compliance. Sunday's New York City brief found 26% of AI and data science postings in the five boroughs omitting the salary range required by Local Law 32. Four years into enforcement, the NYC Commission on Human Rights has filed 396 complaints and never taken a single case to court. The maximum fine is $250,000 per violation. A company running 200 non-compliant AI roles in a quarter carries theoretical exposure of $50 million against which the Commission has not yet tested its ceiling.
Netherlands at 82% voluntary. Sunday's EU bureau documented Amsterdam's rate: 82% of AI roles at Adyen, Booking.com, Philips, and ASML carry salary bands in job postings without any completed national law requiring it. The Netherlands missed the EU Directive's June 7 transposition deadline. Its employers are ahead of the law anyway. Amsterdam at 82% voluntary, NYC at 74% compliance under four years of active fines: that comparison is the arbitrage argument compressed into two numbers.
The pattern across all five data points: regulatory pressure is necessary but not sufficient. Employer culture, talent market competition, and candidate expectations are moving disclosure rates in some markets faster than enforcement is moving them in others. The engineers who have benchmarked against Amsterdam employer offers are now applying those standards when they read a US posting with a $200,000 band spread.
The Global Gap
Today's bureau dispatches map the three largest non-regulated AI hiring markets in ENTRA's coverage universe.
France missed the EU Pay Transparency Directive's June 7 transposition deadline alongside 22 other member states. A senior ML engineer applying for a role at Mistral AI in Paris today sees no legally required salary band. That same engineer, applying for a Mistral role flagged as remote-eligible for a Colorado applicant, would see a posted range. ENTRA's EU bureau monitoring puts voluntary disclosure at fewer than 20% of French AI roles. French transposition is now projected for 2027, per the Ministry of Labour's stated position.
Israel has no salary transparency law and no equivalent directive approaching. The Tel Aviv AI market, anchored by Unit 8200 alumni networks, runs on relationship-level compensation intelligence. Today's ME bureau dispatch maps those comp norms in the absence of mandatory disclosure, including how founders and recruiter networks maintain a proprietary information architecture that candidates outside the network cannot access.
Qatar and the UAE. Gulf sovereign AI employers structure packages as housing allowances, schooling supplements, and gross-up provisions that appear nowhere in any posting. Total compensation for senior ML roles in Abu Dhabi AI clusters runs $180,000 to $320,000 annually; the posting typically shows nothing. Sunday's MOHRE salary architecture briefing and today's Qatar dispatch document what full-package offers look like when the posting is blank.
Together, these three markets represent a disclosure environment that an engineer accustomed to Colorado or Dutch standards would find unrecognizable. The regulatory tide approaching: EU Directive transposition will reach France, Italy, Spain, and Poland through 2026 and 2027 as infringement proceedings move through Brussels. Israel and the Gulf face no equivalent external pressure on the current horizon.
What Changed This Week
Three regulatory signals visible in public data as of August 10 carry direct implications for the pay transparency map.
California's SB 1162 second-cycle expansion. The 2026 reporting cycle expanded California's pay data categories from 10 to 23, with new classifications covering machine learning research, AI safety, model evaluation, and infrastructure engineering as distinct reporting buckets rather than collapsing them into "Computer and Mathematical." Companies that first crossed the 100-employee threshold in 2025 are finalizing inaugural DFEH submissions now. First-failure penalty: $100 per employee, establishing $20,000 in exposure for a 200-person AI startup before compounding years enter the calculation.
Colorado CDLE's shift toward distribution-quality scrutiny. The Colorado Department of Labor and Employment, which logged 2,800-plus complaints and $841,500 in total fines through June 2026, is moving from asking whether a band exists to asking whether the band posted reflects what an employer "actually intends to pay." Multiple 2026 CDLE settlement letters cited band width as a compliance concern. Frontier labs currently posting $180,000 to $220,000 spreads are operating in an enforcement environment that is beginning to examine what those numbers mean for the candidate who reads them.
The EU transposition gap as active arbitrage. Only four of 27 EU member states completed national transposition of Directive 2023/970/EU by the June 7 deadline. That means AI employers with offices in Paris, Warsaw, Madrid, and Rome operate today without the disclosure requirements the Directive would impose. European Commission infringement proceedings against non-transposing states typically run 12 to 24 months. Companies with multi-country footprints are running transparent disclosure in Amsterdam and posting nothing in Milan. That differential is a deliberate compliance architecture, not an oversight.
Five Things to Watch This Week
-
California DFEH inaugural filings. First-time filers under SB 1162 who crossed the 100-employee threshold in 2025 are submitting now. Watch for whether Series B and Series C AI startups appear in the California pay data portal before month's end; absence from the filing list is itself a compliance signal.
-
EU Commission enforcement posture on non-transposing states. The Directive's June 7 deadline passed with 23 of 27 member states still non-compliant. DG EMPL's next scheduled update on infringement referrals comes in September. Any pre-announcement from Brussels this week would reset the timeline for France, Italy, and Spain.
-
The Salary Transparency Act (US Senate). Reintroduced in February 2026, the bill would extend disclosure requirements to all federal postings and establish a private-sector framework overlapping substantially with what Colorado and California already require. Senate HELP Committee markup scheduling for August has not been confirmed. Any scheduling signal this week carries implications for the 2027 legislative timeline.
-
Series B-D startup disclosure behavior. Today's US companion dispatch found 62% of funded AI startups posting no salary range globally, with Series B-to-D companies showing the highest non-compliance rate. Watch for any company in this funding tier adding salary bands proactively during Salary Transparency Month, a pattern ENTRA tracked in August 2025 across 14 companies in the Anthropic alumni cluster.
-
Anthropic's IPO disclosure timeline. Anthropic is reported to have filed confidential IPO paperwork with the SEC in H1 2026 (not confirmed via SEC EDGAR). The first proxy statement will require executive compensation tables that collapse the current information asymmetry between posted bands and realized total compensation. Any movement on the offering timeline this week, registration statement amendments, SEC comment letters, or roadshow signals, marks the beginning of the end of the BBB Pay Transparency rating.
Tuesday publishes the AI Salary Band Quality ranking for the top 50 US employers, scored against ENTRA's disclosure methodology. Wednesday's US analysis covers the Series B-D opacity problem in depth, including which funded companies carry the most exposure under California's 2027 expanded enforcement perimeter. Thursday's industry vertical piece maps the finance-sector AI pay transparency gap, where Goldman Sachs, Citadel, and Bridgewater operate in a compliance universe separate from the Bay Area labs. The companies disclosing least in Week 2 are typically the ones where the information would move the market most. That inverse relationship is the story of the month.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.