New York Tech Week 2026: Women Shaping the Future of AI
As artificial intelligence reshapes industries from healthcare to civic infrastructure, who is making key decisions, and what does responsible AI deployment really look like? A recent event hosted by Fenwick’s Melanie Jolson and Adine Mitrani brought together a cross-sector panel of women influencing the next phase of AI: Dr. Allison Mishkin, child safety and wellbeing product policy lead at OpenAI; Dr. Gal Noyman-Veksler, partner at Lionbird Ventures; and Stacey Matlen, senior vice president of innovation at Partnership for New York City. Read on for takeaways from their conversation.
Grounding AI Safety in Developmental Science, Not Model Cycles
One of the common misconceptions in AI is that safety protocols must be rebuilt every time a model is updated, or a new model is released. In practice, a more durable foundation is the underlying science of human development, which changes far more slowly than technology itself. The science governing how exposure to content shapes young people’s cognition and behavior has remained consistent across decades of research and applies globally, since the neurological mechanisms of child development don’t depend on country or jurisdiction. This framework enables safety teams to write policies that survive model evolutions.
Teams should also consider building AI literacy directly into products, not just communicating it externally. Tools that nudge users toward better prompting habits, flag potentially risky use patterns in real time, or surface proactive guidance while someone is actively engaging with a model can raise the overall quality of interaction without requiring users to seek out separate education or review policies that live outside of the platforms themselves. The ease of use that makes generative AI powerful may also create a false sense of confidence, making in-product guidance particularly valuable.
How Cities Are Testing and Scaling AI: A Structured, Problem-First Approach
For government agencies, the most successful AI deployments sometimes begin not with a technology selection but with a disciplined problem definition. Speaking with the frontline staff who deal with a challenge daily (not only with executive leadership) may help produce a more accurate picture of what a solution actually needs to do. Layering in top-down prioritization then ensures that selected problem areas have both operational legitimacy and institutional support.
A structured evaluation process matters as well. Operators who understand the problem may be best suited to review applications, hear pitch presentations, and assess which technologies are likely to make a difference in their day-to-day work. Breaking up pilots into iterative phases, beginning with a proof of concept using dummy or public data, followed by a longer pilot in the real operating environment, may increase the signal quality of each test.
The AI solutions that successfully scale in public sector environments tend to deliver against one of three measurable outcomes: efficiency, resilience, or responsiveness. A clear, quantifiable connection between the technology and a problem that needs to be solved is key.
AI in Healthcare: Domain Expertise, Human Oversight, and the Equity Gap
Early-stage healthcare AI investors frequently encounter founding teams before any product exists, which means diligence is based on the people and their abilities as much as the technology. An important signal is whether the founding team understands the regulatory environment, clinical constraints, and buyer behavior specific to their segment. Deep domain expertise (whether through a full-time clinician, a fractional advisor, or a research partnership with an academic institution) functions as both a quality check on product direction and a credibility signal to sophisticated buyers.
On the safety side, a key structural principle is keeping the human in the loop. Healthcare AI that augments clinicians rather than replacing them, and that delivers recommendations rather than autonomous decisions, may be more defensible from a safety, regulatory, and liability standpoint. Products built alongside psychologists and leading clinical researchers are better positioned to handle the edge cases that pure AI optimization tends to miss.
A significant equity dimension is emerging in the healthcare AI space: Women were excluded from clinical trials until the mid-1990s, leaving a substantial gap in sex-disaggregated data. AI tools now give pharmaceutical and clinical companies the ability to analyze subpopulations with far greater granularity, creating a statistical opportunity to address decades of underrepresentation. Founders building in this space should recognize that female-dominated conditions (autoimmune diseases, hormonal conditions, and cardiovascular disease) represent enormous markets. Autoimmune conditions alone carry an estimated $100 billion in annual spend, with women comprising approximately 80% of patients. Investor interest in women’s health AI reflects this market logic, with women’s health capturing an estimated 25 to 30% of health AI funding in the first quarter of this year.
Closing the Gender Gap in AI Adoption
Research consistently shows that women are less likely to adopt AI tools than men. One widely cited study found a 16-percentage-point gap in usage. When asked why they were not using AI, women most commonly cited a need for training before they felt ready. Men, by contrast, cited employer restrictions, and many used AI anyway. The implication is not that women are less capable or less interested, but that the pathway to adoption is different and that the current framing of AI tools is not working for a significant portion of potential users.
The positioning of AI tools skews heavily toward language associated with engineering and technical manipulation: hacking, coding, and building. Reframing AI as a problem-solving tool, one that reduces the time it takes to tackle complex challenges, might resonate more broadly, particularly with professionals whose instinct is to lead with the problem rather than the technology. Women have consistently demonstrated strong problem-solving orientation across public health, civic design, and entrepreneurship, and tools framed as supporting that focus are more likely to be adopted.
Making Safety a Business Priority
Safety concerns in AI (whether around child protection, mental health, or healthcare) are frequently coded as soft issues rather than core product requirements and changing that perception may require quantification. Framing a child safety risk in terms of the percentage of users affected and the likely developmental outcomes or framing a healthcare risk in terms of regulatory exposure and reimbursement risk translates the concern into the language that engineering, legal, and commercial teams may respond to.
For founders and early-stage companies, it may not be sufficient to demonstrate value alone. Rather, buyers (whether enterprises or government agencies) expect a clear account of how a product was validated, how it is monitored, and what happens when it fails. Having a plan for when things go wrong is as important as having a plan to prevent them from going wrong in the first place.
Strong legal counsel is also critical. Privacy missteps, data governance failures, and inadequate safety documentation may invalidate a company’s core value proposition even after it has achieved market traction. Tiering risk by use case; recognizing that AI serving children, patients, or insurance underwriting carries materially different exposure than back-office automation; and building a legal infrastructure accordingly may allow early-stage companies to manage risk without stalling product development.