Meet the Young Innovators Building Guardrails and Breakthroughs Into Tomorrow's Tech

Every technology we build carries hidden costs. Sometimes it's electricity wasted as heat. Sometimes it's a financial app that unintentionally hands an abuser a backdoor into a victim's bank account. The researchers profiled here aren't policymakers. But their work is quietly redefining what responsible technology design looks like, and regulators would do well to pay attention.

Take the energy problem first, because it's the one hiding in plain sight inside every data center on the planet. Hannah Earley, 31, is CTO and cofounder of Vaire Computing. She's tackling a basic inefficiency in how computers work. Traditional chips waste much of the energy they consume, dissipating it as heat that then has to be cooled, often by water-hungry systems straining local grids.

Earley's answer is reversible computing. Think of it like recirculating water in a fountain instead of draining it after a single use. Rather than discarding information (and the energy tied up in it) once a calculation is done, a reversible computer reverses the process to recover that energy. In 2025, her team unveiled a proof-of-concept circuit that achieved net energy recovery, meaning it saved at least as much energy as it cost to recover. "The physics promises that this is the best way to build a computer," Earley says. If the approach scales into real manufacturing, it could help decouple AI's growth from ever-rising electricity demand, a policy concern that grows more urgent as data centers multiply.

Designing for Harm Reduction

Energy is one kind of hidden cost. Financial abuse is another, and it's far more personal.

Rosanna Bellini, 32, an assistant professor at NYU, has spent years documenting how banking apps and payment platforms can become tools of intimate partner abuse. After completing her PhD at Newcastle University, she moved to Cornell Tech in 2021, where she worked directly with more than 120 domestic violence victims through Cornell's Clinic to End Tech Abuse, plus roughly 80 more through other programs.

That hands-on experience led her to a harder question: why do abusers do this in the first place? She interviewed hundreds of abusers enrolled in behavior-change programs and combed through online forums, compiling more than 500 firsthand accounts of people who had surveilled partners through spyware or other means.

Her most consequential work may be the audit she and collaborators conducted of 30 consumer finance apps, including tools from JPMorganChase, TD Bank, PayPal, Cash App, and Venmo. It was the first study of its kind. The team found design flaws that let abusers gain covert access to a partner's account with no technical skill required. In Chase's app, they discovered a way to add a second fingerprint or biometric login without ever notifying the account holder. After Bellini's team flagged it, JPMorganChase updated the app to alert users whenever a new biometric ID is added.

"We can actually really make a huge difference in the design of these systems," she says. Small fixes, she notes, can prevent real trauma.

Other researchers are approaching risk and access from entirely different angles, but the throughline is the same: building systems that anticipate how people will actually use, or misuse, them.

Ariel Ekblaw, 34, who leads the nonprofit Aurelia Institute and cofounded the company Rendezvous Robotics, is rethinking how we build infrastructure in space. Current construction methods rely on astronauts assembling components by hand in bulky suits, work that is slow and genuinely dangerous. Ekblaw designed hexagonal and pentagonal tiles with embedded magnets that self-assemble once released in orbit, cutting out much of the risky manual labor. The approach has already been tested twice on the International Space Station using small models, with a larger 32-tile demonstration planned for later this year and an exterior test expected in 2027.

Shuang Li, 32, a research scientist at Google DeepMind, is working on a different kind of reliability problem: getting robots to function outside the narrow conditions they were trained in. During her PhD at MIT, she contributed to early research showing language models could guide robotic decision-making. At Stanford, she led development of Unified Video Action, an open-source training model released in 2025 that teaches robots to track how their actions change what they see, rather than getting distracted by unfamiliar backgrounds.

Jianlan Luo, 33, at the Shanghai Innovation Institute, is speeding up how robots learn new tasks altogether. His research showed that incorporating human feedback during training doubled robots' success rates on tasks like flipping an egg or assembling furniture. Back in China, he's now built a framework, tested with humanoid robot maker AgiBot, that lets robots upload performance data to a shared database so an entire fleet improves together. He's scaled the method across 16 robots so far and hopes to test it on 100.

Jelena Notaros, 33, at MIT, is shrinking optical systems down to the nanoscale using silicon photonics, the same lithography techniques already used to manufacture computer chips. Her work has produced a quarter-sized 3D printer, an optical tweezer capable of manipulating mouse cells, and a prototype for lidar sensors that could shrink to the size of an uncooked lentil while detecting objects at greater distances than today's spinning rooftop units on autonomous vehicles.

And Patrick Slade, 32, at Harvard, is addressing a persistent access problem in assistive technology: one in six people worldwide live with mobility disorders, yet customized devices remain expensive and slow to produce. His human-in-the-loop optimization approach lets an ankle exoskeleton be trained to an individual's needs in just 20 to 40 minutes of normal use, reducing walking energy expenditure by 23 percent in testing while also increasing speed.

Why It Matters

None of these researchers set out to write policy. But their work touches exactly the questions regulators are grappling with: how much energy AI infrastructure should be allowed to consume, how financial products should be audited for hidden harms, and how quickly assistive technology can be made affordable and personal. The lesson across all of it is consistent. Build the safeguard into the design early, and you spare people, and sometimes the planet, from paying for it later.