62% of Series B-through-D funded AI startups list no salary range on their public careers pages, per the ENTRA Job Signal Index Q2 2026 analysis of 847 US-based AI company job postings (a US sub-sample of the Index's 12,000-plus global dataset; both the US cohort and the full global set independently produce the 62% rate). That figure sits alongside a well-documented benchmark from the frontier lab tier: Anthropic posts $120K-$320K in base salary bands; OpenAI has moved to structured range disclosure across the majority of its US listings; xAI began attaching ranges to California and Colorado postings in Q1 2026. The transparency gap between those frontier-lab disclosures and the mid-market AI tier, companies operating at $50M-$500M ARR with Series B, C, or D capital, is not narrowing. Three US laws now mandate salary disclosure in the states where most American AI talent concentrates. The companies that have read those laws carefully enough to find their limits are the ones keeping their ranges hidden.
The Three Laws and the Gaps They Leave
Colorado's Equal Pay for Equal Work Act set the template for US pay range disclosure when it took effect in January 2021. Its 2025 amendments sharpened the rule most relevant to remote-first AI startups: any employer with at least one Colorado-based employee must post compensation ranges on all job listings for roles that employee might perform, regardless of where the company is headquartered. Fines run $500 to $10,000 per violation. A San Francisco-based Series C AI company with a single remote engineer in Denver is a Colorado employer for every role that engineer could theoretically fill.
California's SB 1162, which requires employers with 15 or more California-based employees to post pay scales on all job listings, added mandatory penalties through SB 464 (effective January 2026). The California Civil Rights Department can now levy $100 per employee per posting violation, with no "good faith" exemption available. For a 200-person startup with 60 California employees running 40 open listings without salary disclosure, the statutory exposure is $2.4 million in aggregate, not a theoretical ceiling that enforcement will politely ignore. As previously reported in this series, California's first SB 1162 annual filing cycle under mandatory penalties closed in August 2026. The CRD is expected to begin targeted enforcement letters to non-compliant tech employers in Q3 2026.
NYC Local Law 32, the third leg of this regulatory stool, applies to employers with four or more employees and reaches any role that could be performed in New York City, including remote listings that name NYC as a permissible work location. Per-violation exposure reaches $250,000. Our August 9 coverage detailed the enforcement posture; the operative point for mid-market AI companies is that NYC's four-employee threshold is low enough to capture virtually every funded startup operating in the US.
The gap that 62% of mid-market AI startups are exploiting is not ambiguity within Colorado, California, or New York. Those three jurisdictions are specific. The gap is every other state. A startup headquartered in Austin that lists roles as remote but explicitly excludes Colorado, California, and New York from eligible work locations is technically exempt from all three disclosure regimes. Two HR operators at Series C AI startups, granted anonymity in ENTRA Q2 sourcing conversations, described this approach as "geographic compliance scoping," a deliberate eligibility-restriction policy. Among ENTRA's Q2 sample of 847 postings, approximately 19% of remote listings carried explicit state exclusions for at least one of the three regulated jurisdictions. That number was 6% in Q2 2024.
Who Posts, Who Hides
The mid-market AI tier is not uniform on disclosure. A few companies have made proactive posting standard. Several are in partial compliance. A meaningful portion has made opacity a deliberate policy.
Glean, the enterprise AI search platform that closed at a $4.6B valuation and now operates near the top of the mid-market AI tier, posts salary ranges on a majority of its engineering listings. A Machine Learning Engineer, AI Assistant and Autonomous AI Agents listing active on Glean's Greenhouse board in Q2 2026 carried a disclosed base of $240,000-$300,000. That number is visible before the first recruiter call. Glean's California employee base makes SB 1162 compliance a threshold obligation, and the company appears to be meeting it consistently.
Harvey, the legal AI company that raised $300M at a $3B valuation in 2024, presents a different picture. Levels.fyi aggregates through mid-2026 place Harvey software engineer median total compensation at $390K-$492K. The company's Greenhouse job listings, reviewed by ENTRA through Q2 2026, do not consistently include salary ranges on non-California postings. A candidate applying to a Harvey "remote, US" role without California residence encounters no posted range, no stated band, and no indication of where inside a $100K-plus spread the offer will land.
Perplexity AI, which raised $500M at an $8B valuation in April 2024 and extended to a $2.8B round in June 2025, operates in similar fashion. Market data from Levels.fyi and 6figr places Perplexity AI engineer total compensation at $352K-$776K across experience bands. Active job listings do not consistently include salary ranges. Perplexity's San Francisco headquarters triggers California SB 1162 obligations for listings that identify California as a work location, but remote-eligible listings that omit a state designation remain dark.
Cohere, headquartered in Toronto with a US employee base concentrated in New York and San Francisco, presents a fragmented disclosure picture. Glassdoor salary data for Cohere spans approximately $82K-$510K depending on role and reporting source, a spread that reflects the absence of anchor data from posted bands. For a company competing against Anthropic for the same enterprise AI sales and applied research profiles in New York and California, the information asymmetry lands hardest on the candidates who need it most.
Cognition AI, known for the Devin coding agent and valued at $2B after its 2024 Series B, and Writer, the enterprise AI platform that raised $200M in a 2024 Series C, both operate careers pages where the majority of postings reviewed by ENTRA through Q2 2026 carry no salary disclosure. Neither company has published a compensation philosophy or a pay transparency statement. Character.AI, which transferred its founding team to Google in a licensing structure in 2024 before the remaining entity raised additional capital independently, has shifted portions of its hiring under Google's band infrastructure, but the residual Character.AI brand entity's independent postings carry no consistent disclosure.
The pattern across this tier is not random. Companies with higher concentrations of California and New York employees, like Glean, tend toward disclosure because the cost of non-compliance is calculable and large. Companies with more geographically distributed remote workforces, or with HR teams that have invested in eligibility-restriction architecture, tend toward opacity. The choice is a compliance calculation, not a culture signal.
What Candidates Do When the Bands Are Missing
When salary ranges are absent, candidates in the mid-market AI hiring market have developed three parallel sourcing habits, and the data suggests those habits are partially closing the information gap that non-disclosing companies are counting on.
Levels.fyi is the first lookup. The platform's company-specific pages aggregate self-reported compensation data by role, level, and location. For Harvey, Perplexity, Glean, and Cohere specifically, Levels.fyi has accumulated enough data points that a candidate can construct a credible band estimate before the first recruiter call. Per a June 2026 analysis by LyncMe, Levels.fyi is now used as a pre-call preparation tool by the majority of senior AI candidates, not as a post-offer validation check. That shift compresses the information gap, but it does not eliminate it: Levels.fyi data lags four to eight weeks behind real-time market moves, and sample sizes for companies under 500 employees are thin.
Glassdoor and Blind serve different functions. Glassdoor surfaces lower-range estimates with more noise, but its review data frequently includes recruiter behavior signals: whether initial offers landed near the ceiling or the floor, how often the company resisted negotiation, what the delta between first offer and accepted offer typically runs. Blind's AI startup compensation threads, updated within weeks of new offers at Harvey, Glean, and Perplexity, are now standard pre-call research for candidates at Staff level and above.
The third behavior is the most commercially meaningful: candidates with a posted-range offer from any disclosing company, including Anthropic's $120K-$320K base band, use that posted figure as a negotiation anchor when speaking to non-disclosing companies. The Anthropic band understates actual L5 total compensation by 95%, as our August 9 analysis found. But for a candidate entering a first Harvey or Perplexity offer conversation, the fact that a named frontier lab has put $320K in writing gives them a floor to defend. ENTRA recruiter network data from H1 2026 (n=214 placements across mid-market AI companies) shows that candidates who arrive at a non-disclosing company's recruiter call with Levels.fyi data in hand close offers that are on average 14% higher than first offers. The gap is partly negotiation skill. Part of it is simply information.
The Q4 2026 Enforcement Trajectory
The pressure on mid-market AI startups is not peaking in August 2026. Twelve US states now require salary ranges in job postings. Illinois joined the list effective January 2025. Minnesota's law took effect the same month. Massachusetts' pay transparency statute, covering employers with 25 or more employees, enters active enforcement for the 2025 reporting year in fall 2026. New Jersey's law, covering employers with ten or more employees, adds another AI-employer-dense jurisdiction by Q4.
The arithmetic is unfavorable for the "geographic compliance scoping" strategy. A mid-market AI startup with 25 employees and any remote presence across IL, MN, MA, NJ, CO, CA, and NYC is either disclosing on essentially all its postings or carrying enforcement exposure across five or more jurisdictions simultaneously. The eligibility-restriction playbook that worked in 2024 stops working as the map of covered states covers the ZIP codes where AI candidates actually live.
The ENTRA Job Signal Index will re-survey the 847 postings examined in Q2 in November 2026. The directional read: the 62% non-disclosure rate drops below 50% by year-end, not because mid-market AI companies have decided transparency is a recruiting asset, but because the statutory map has made opacity arithmetically expensive. The companies that lead on disclosure before that deadline get the brand benefit of having chosen it. The ones that wait will post bands under regulatory pressure, with no credit for timing.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.