AI

Under-35 Innovators Transform Biotechnology and Steel Decarbonization

MIT names innovators under 35. Vision restoration, gene editing, generative AI design, and low-emission steel furnaces explained.

7 min read Reviewed & edited by the SINGULISM Editorial Team

Under-35 Innovators Transform Biotechnology and Steel Decarbonization
Photo by Julia Koblitz on Unsplash

MIT Technology Review has published the 2026 edition of 35 Innovators Under 35. It is the annual feature selecting groundbreaking achievements from young innovators under 35 around the world. This year, selections were made in four categories: artificial intelligence and computing, biotechnology, and climate and energy. In reporting by Thomas Macaulay of MIT Technology Review AI, the focus is on nine people in the biotechnology category and an entrepreneur tackling decarbonization of steelmaking. The September 11, 2026 edition of the weekday newsletter The Download summarized the highlights.

In the biotechnology category, vision restoration, rare-disease treatments, brain measurement, and virus design with generative AI were featured. In the climate and energy category, a new furnace that produces liquid steel from iron ore in a single step drew attention. At the same time, blocking AI misuse and legislation to protect minors were listed as key reads. All go beyond laboratory results, sharing the feature of being directly linked to social implementation and safety measures.

Young Innovators Take On Vision Restoration

and Gene Editing

The nine in the biotechnology category are pushing existing limits on both treatment and measurement. A representative example is a “reprogramming” therapy that reverses vision loss. It aims to restore visual function by reverting cells toward an earlier state. It offers a new option for vision loss from retinal disease long considered irreversible.

Tiny brain electrodes inspired by Japanese art were also selected. They apply deployable structures rooted in origami and kirigami to reduce the burden on brain tissue. Designed to flexibly deform and adhere closely, they aim to stabilize signal acquisition. The effort is positioned as balancing higher precision with less invasiveness in implantable measurement.

Also included was a personalized gene-editing treatment for an infant with a rare genetic disorder. Its distinctive feature is tailoring the editing approach to the patient’s unique mutation. It requires a different development and review framework from mass-produced pharmaceuticals. The fact of establishing a treatment for a single individual marks a milestone for personalized medicine.

“tiny brain electrodes inspired by Japanese art, and a personalized gene-editing treatment for a baby with a rare genetic disorder”

The above is part of the selection examples cited in the original report. Challenges in longevity technology are also included, bringing practical application of aging interventions into view. Young-led ideas are reaching clinical application ever faster. The boundary between research and entrepreneurship is becoming even thinner.

Generative AI Virus Design Efforts

An attempt to design novel viruses with generative AI also drew attention. It is research aimed at creating pharmaceuticals and absorbing pollutants. It builds on advances in protein structure prediction and sequence generation. The workflow computationally explores target binding and stability, then validates experimentally.

The original report touches on the potential with the caveat of “if it works.” While applications to new drugs and environmental remediation are expected, ensuring biosafety is a prerequisite. Assessing the impact of engineered viruses on ecosystems and the human body is essential. Strict enforcement of containment, sequence screening, and experimental approvals is required.

This field symbolizes the acceleration of biotechnology by AI. Because the search space is vast, the contribution of generative models is large. At the same time, concerns about misuse and leakage are ever-present. Designs that reconcile innovation with control are called for.

Steel Furnace Switching From Coal to Natural Gas

Steelmaking has changed little in its basic process since commercialization in the 1850s. Iron ore smelting has been characterized by coal dependence and multi-stage processing. Laureen Meroueh, founder of Hertha Metals, has taken on this structure. She developed a new furnace that produces refined liquid steel from iron ore in a single step.

Another feature is replacing coal with natural gas as fuel. According to her, it can cut emissions by at least half. She says it can reduce costs by 25% compared with conventional steelmaking. The design aims to achieve both decarbonization and lower prices.

“Her new furnace turns iron ore into refined liquid steel in a single step and swaps coal for natural gas”

Consolidating processes leads to lower equipment costs and heat losses. Fuel switching directly curbs carbon dioxide emissions. Steel is a foundational material accounting for a large share of global emissions. If the single-step process is proven, pressure to replace existing blast furnaces will intensify.

Challenges lie in feedstock compatibility and quality stability at scale. Variations in iron ore grade and impurity control affect yields. Natural gas procurement prices and supply networks also affect profitability. Accumulating operational furnace data and third-party verification will be conditions for adoption.

Anthropic Details Efforts to Block AI Misuse

Anthropic disclosed that it blocked attempts that could have led to bioweapon development. Five cases were confirmed. There were also six cases in which AI was used to support program development for conventional weapons. The company’s new report shows the reality of the threats.

Surveillance use of Claude by governments was also reported. A case was cited in which Russia-linked attackers used it to automate attacks on Ukraine. It was allegedly misused to generate attack procedures and streamline target research. Detection by defenders and blocking under terms of use are central to the response.

Rights disputes over AI are also intensifying. As shown in Round Hill’s $1B+ Lawsuit Against Suno and Anthropic, the boundary between training use and copyright is being contested in court. Safety measures and rights clearance are being institutionalized in parallel. In tracking attack automation, the increasingly sophisticated investigative techniques covered in Windows GDID Helps Identify Scattered Spider Suspects are also instructive. Operators are being pressed to strengthen systems for misuse detection and evidence preservation.

California’s Anti-Addiction Regulation

California has banned addictive features aimed at those under 16. Infinite scroll and autoplay are prohibited. New rules on AI and interactive companion chatbots were also introduced. It is the first law of its kind in the United States.

The aim is to curb designs that encourage extended usage time. Mechanisms that automatically display one item after another extend time spent. Concerns have been raised that they could undermine minors’ autonomous usage decisions. The state has shifted toward requiring age-appropriate considerations at the design stage.

Covered businesses will need to change defaults for recommendations and playback. Implementing age verification and parental controls will also be a challenge. Spillover to other states and consistency with federal law will be future focal points. The balance between platforms’ design freedom and user protection is at issue.

Editorial Opinion

Looking at short-term impacts. In the next 3-6 months, review procedures for personalized gene editing will draw attention. The infant case will prompt debate over cost burden and expedited approvals. In steelmaking, disclosure of demonstration data for the new furnace will progress. Reproducibility of halved emissions and 25% cost reduction will be tested. In AI safety, screening of biology-related uses will be strengthened. Companies will be asked to make detection criteria more transparent.

Taking a long-term view. In 1-3 years, generative-AI virus design will be integrated into drug-discovery platforms. International standards for sequence screening and containment will be developed. In steelmaking, the single-step process will affect existing facility renewal plans. The transitional approach via natural gas will become a bridge technology toward electrification. Minor-protection regulations will spread to other regions as design norms. Safety by default will become a competitive condition.

We note questions from the editors. How society should share the costs of personalized treatments is in question. Evaluation methods when benefits are limited to one person are also unsettled. Who certifies emission reductions for low-emission steel is also an issue. The line between blocking AI misuse and surveillance use remains blurred. Which to prioritize, safety or convenience, is being asked.

References

Frequently Asked Questions

What technologies stood out in the biotechnology category of 35 Innovators Under 35
Reprogramming therapy reversing vision loss, tiny brain electrodes inspired by Japanese art, and personalized gene editing for an infant with a rare disease were cited. Novel virus design with generative AI and longevity technologies are also included. Balancing clinical application with safety assessment is the hallmark.
How is Hertha Metals’ new furnace different from conventional steelmaking
It produces refined liquid steel from iron ore in a single step and replaces coal with natural gas as fuel. According to the founder, it can cut emissions by at least half and reduce costs by 25%. Feedstock compatibility, quality stability at scale, and third-party verification will be key to adoption.
What do the moves by Anthropic and California show
Anthropic reported blocking misuse in 5 bioweapon-related cases and 6 cases of program support for conventional weapons. Government surveillance use and attack automation were also confirmed. California banned infinite scroll and similar features for those under 16. Institutionalization of safety and rights protection is advancing.
Source: MIT Technology Review AI

Comments

← Back to Home