The Law Meant to Break Silence—and the Hiring Signal It Sent
The subject of this series is the response of U.S. venture-capital-backed startups to laws that weaken harassment-related nondisclosure agreements. The immediate question is unsettling: why would a reform intended to help people speak about sexual harassment be followed by fewer women entering the affected firms? The answer matters because it reveals how a legal right can change an organization’s incentives before the organization has changed its culture. By the end of this series, the useful judgment is not whether transparency is simply good or bad, but whether a startup has the capacity to turn transparency into credible prevention and accountability.
The paradox belongs to the startup labor market
A recent NBER working paper studies state laws adopted between 2018 and 2022 that weakened the enforceability of NDAs in sexual-harassment cases. Using employment histories linked to records for more than 50,000 U.S. venture-backed startups from 2014 through 2022, the authors report that treated startups hired about 8% fewer women per year after their state’s reform. They find no meaningful change in male hiring. The reduction is therefore not merely a story about young companies shrinking their workforces; it is a gender-specific change in whom they bring in.
That result runs against the moral direction of the policy. Silence around harassment is not a neutral absence of information. It can keep a victim from reporting, conceal patterns from other workers, and protect people with power from reputational or legal consequences. Congress made that logic explicit in the federal Speak Out Act, enacted in December 2022: for covered sexual-assault and sexual-harassment disputes, a nondisclosure or nondisparagement clause agreed to before the dispute arises is not judicially enforceable when the alleged conduct violates applicable law. The statute’s stated purpose is to give survivors more freedom to report and publicly disclose abuse, not to preserve an organization’s ability to make misconduct disappear.
The study’s short-run estimate does not overturn that purpose. It identifies a possible cost of changing the information environment. Once a firm can no longer rely on a harassment-related predispute promise to keep a future dispute quiet, the expected consequences of employing people in relationships where harassment might occur become more visible. A startup may face a greater chance of litigation, public scrutiny, investor concern, or employee flight. Those risks can be priced into decisions made before anyone has filed a complaint.
This is the key distinction: a law can be protective for a worker and still change an employer’s demand for labor. The two statements are not logically incompatible. A worker may gain a stronger ability to speak, while a firm—especially a poorly prepared firm—responds by trying to reduce the situations in which it expects that right to be used. The NBER finding is evidence of that tension in a particular population of firms, not proof that the policy was designed badly or that every startup behaves this way.
The setting matters. A venture-backed startup is usually young, small, and still establishing its hierarchy. A first cohort of employees does more than fill open roles: it creates referral networks, informal norms, and the model of who is expected to have authority. The average startup in the study is only a few years old and hires a small number of people each year. If the firm delays or avoids its first female hires, the numerical effect in one year can become a cultural effect through the composition of the people who later recruit, manage, and mentor others.
At the same time, a small startup may have no dedicated HR staff, independent reporting channel, or mature compliance process. That does not mean every small firm is unsafe, or that every larger firm has solved harassment. It means size can stand in for a practical constraint: when the rule changes, some firms have fewer established ways to prevent misconduct, investigate a report, protect a worker, or demonstrate to investors that the problem is being managed. If the only familiar tool was confidentiality, weakening that tool can feel like exposure without an alternative system.
The missing distinction is not “NDAs: good or bad”
An NDA has a legitimate use in a technology startup. Its original function is to protect trade secrets, proprietary information, and other knowledge that a firm needs to share with employees without handing it to competitors. The federal law preserves that function. It does not say that every confidentiality promise is void, nor that a company must publicly disclose its intellectual property or every internal disagreement.
The controversy begins when the same contractual architecture is used to restrict discussion of harassment or assault. The startup study distinguishes two moments. A predispute NDA is presented as part of employment before anything has happened; it can shape what a worker believes they may say if a problem arises. A post-dispute NDA may be signed as part of a settlement, binding a person to confidentiality after an allegation or claim. The policy wave does not erase the difference between these arrangements, and state laws vary in which arrangements they target. But all of them raise the same institutional question: when a worker knows that disclosure may be restricted, who else can learn that a pattern exists?
That question helps explain why the policy can affect hiring without requiring a harassment event at every treated startup. The law changes the firm’s exposure ex ante. Before the reform, a manager may have imagined a complaint as an event that could be contained through an employment clause or settlement. After the reform, that containment is less reliable for covered disputes. A rational firm might respond by investing in prevention and reporting systems. It might also respond defensively by avoiding hires it associates with greater exposure. The study’s pattern suggests that the second response appears first in firms with weaker apparent capacity.
This is where the central theme of the series comes into view: a transparency-capacity gap. Transparency is the external condition that makes misconduct more knowable. Capacity is the internal ability to act on what becomes knowable. If transparency arrives first, the firm may experience the change as a new liability. If capacity is already present, the same change can be absorbed through clear behavioral expectations, independent reporting, prompt investigation, support for the person who reported, and action that makes future harm less likely.
The distinction also prevents an overly simple reading of the 8% result. The study does not show that women are the source of a problem, that a woman’s presence causes harassment, or that firms consciously apply a written “risk discount” to every female candidate. It observes aggregate hiring patterns after a legal change and tests which patterns are consistent with a demand-side response. Nor does it show that allowing disclosure creates harassment. The legal theory behind the reform is that silence can allow existing misconduct to persist; the empirical question is how organizations react when silence becomes less dependable.
The broader workplace evidence points to why the replacement system matters. In a qualitative PLOS One study, workers who had experienced harassment described clear leadership messages before an incident as one form of prevention. When harassment occurred, they valued a reporting option outside the immediate chain of command, verbal acknowledgment that the report was serious, and follow-up action that protected the target and addressed the conduct. These findings come from 30 interviews, mostly in one region and largely in service work; they are not a statistical description of startups. Their value here is more specific. They show what “capacity” looks like from the worker’s side. A policy can make speech legally safer while a workplace still makes speaking practically dangerous if the only reporting route leads back to the person with power.
Recent research summarized in an abstract from Human Resource Management makes a related conceptual point: organizational silence can involve being silent, silencing others, and not listening. A clause is only one possible signal. A manager who dismisses a complaint, colleagues who punish a reporter socially, or an institution that never explains what it did can all communicate that disclosure will not produce protection. Weakening an NDA addresses one formal barrier; it does not automatically remove those informal ones.
The diagram is not a claim that every firm follows one fixed sequence. It is a map of the relationship the evidence makes worth examining: a formal change in disclosure rights can move through expected exposure into an external hiring decision, while the same pressure may also produce internal changes. Capacity determines which response is more feasible. The long-run endpoint remains a question, not an arrow the data have already validated.
Two clocks, one unresolved question
The first clock is immediate. The legal rule changes, firms reassess exposure, and hiring can move within a year. The NBER study reports that the fall appears sharply after enactment and persists through its state-level sample. The effect is concentrated among junior women and in small or male-dominated startups—precisely the settings where power differences may be greater and internal safeguards weaker. That pattern gives the result an organizational shape; it is not just a before-and-after headline.
The second clock is slower. The same study finds more women promoted to managerial positions and more male managers departing after reform. Some departing managers later experience lower pay, lower seniority, or longer unemployment, which is consistent with at least some involuntary separations. But the data cannot establish that every departure removed a harasser, or even distinguish cleanly between employer action and men withdrawing from relationships they now perceive as risky. The firms may be changing their leadership, but that does not yet tell us whether workers experience the change as safer, fairer, or more supportive.
This creates the real issue for the series. If a startup responds only by reducing female entry, the policy’s protective purpose is undercut at the labor-market margin. If it also builds credible systems, changes who holds authority, and makes accountability routine, the initial hiring contraction might eventually give way to a healthier workplace. The evidence does not yet show that recovery. The state analysis ends in 2022, and the federal comparison reaches only through 2024; the authors explicitly treat long-run adaptation as an open question.
So the next question is more precise than whether the law “worked.” What, exactly, became unenforceable when harassment-related NDAs were weakened, and what remained available to a startup that still needed to protect legitimate confidential information? Until that legal and contractual boundary is clear, the hiring signal is easy to misunderstand—and the difference between transparency and culture remains invisible.