The Download: Biotechnology’s Next Generation and a Cheaper, Cleaner Route to Steel

The September 11, 2026, edition of The Download brings together two very different areas of technological change: biotechnology, where younger researchers are pursuing new treatments and biological tools, and steelmaking, where an alternative furnace design could reduce emissions without raising costs.

The newsletter also highlights concern about the misuse of artificial intelligence, new restrictions on social-media design for children in California, departures by AI-safety researchers, efforts to protect critical infrastructure, and the music industry’s changing relationship with generative AI.

Biotechnology’s rising innovators

MIT Technology Review’s 2026 35 Innovators Under 35 includes nine people working in biotechnology. Their projects span medical treatment, longevity research, genetic engineering, neuroscience, and environmental applications.

The examples highlighted in the newsletter include:

• A reprogramming therapy intended to reverse vision loss.
• Very small brain electrodes whose design was inspired by Japanese art.
• A personalized gene-editing treatment developed for a baby with a rare genetic disorder.
• The use of generative AI to design new viruses, with possible applications in drug development or pollution removal.

These projects represent the breadth of current biotechnology. Some aim to repair or modify the human body directly; others use biological systems as tools for manufacturing, medicine, or environmental cleanup.

Why these technologies are difficult to assess

The promise of biotechnology often arrives before its practical limits are fully known. A treatment that works in an early setting may still face questions about safety, durability, manufacturing, regulation, and access. Gene editing and biological design also raise risks that do not apply to many conventional software tools.

The newsletter’s mention of AI-designed viruses illustrates that dual-use problem. The same techniques that might help researchers develop drugs or absorb pollutants could, if misapplied, support harmful biological activity. That makes technical progress inseparable from safeguards, testing, and oversight.

The honorees are part of a broader 2026 list covering artificial intelligence, computing and robotics, biotechnology, and climate and energy. Recognition does not establish that a technology is ready for widespread use.

A proposed route to cheaper, cleaner steel

Steelmaking has changed relatively little in the core process used to purify iron ore since the nineteenth century. Laureen Meroueh, founder of Hertha Metals and one of the climate and energy honorees, is pursuing a different approach.

Her company’s furnace is designed to:

1. Convert iron ore into refined liquid steel in a single step.
2. Replace coal with natural gas.
3. Reduce emissions by at least half, according to Meroueh.
4. Cut costs by 25% compared with conventional steelmaking, also according to her.

The proposition is significant because decarbonizing heavy industry often involves a trade-off between emissions and expense. A process that reduces pollution but makes steel substantially more costly can face resistance from manufacturers and buyers. Hertha Metals is instead presenting cleaner steel as potentially cheaper steel.

What the proposed process changes

Traditional steelmaking involves multiple stages and relies heavily on coal. A single-step furnace could reduce the number of processing steps, while using natural gas instead of coal could lower emissions from the fuel itself.

That does not make the process carbon-free. Natural gas is still a fossil fuel, and the total climate impact would depend on factors such as the furnace’s energy efficiency, the source and handling of the gas, and the emissions associated with mining and transporting raw materials.

The figures cited in the newsletter are claims from Meroueh, not independent results presented in the source material. The commercial test will be whether the process can maintain those cost and emissions advantages at industrial scale while producing steel that meets demanding quality requirements.

AI misuse moves from speculation to reported cases

Anthropic said it had blocked five potential attempts to use its systems in the development of biological weapons. The roundup also mentions six cases involving the use of AI to build software for conventional weapons.

These disclosures point to a more concrete form of AI risk: not merely the possibility that advanced systems could be misused, but the need for companies to identify and interrupt suspicious activity in practice.

The cases described by Anthropic do not establish that a biological weapon was built or deployed. They indicate that the company detected activity it considered sufficiently concerning to block. That distinction matters when interpreting claims about AI-enabled threats.

The roundup also reports that:

• Governments are using Claude for surveillance.
• Russia-linked hackers used the system to automate attacks on Ukraine.
• Anthropic documented these and related concerns in a new report.
• Bill Gates has argued that AI requires additional guardrails.

Together, the reports show that the same general-purpose systems can be used across very different settings, from defensive or administrative work to surveillance, cyberattacks, and weapons-related activity. Blocking individual cases is therefore only one part of the broader safety question. The effectiveness of safeguards also depends on how systems are monitored, how suspicious users are identified, and whether companies disclose enough information for outsiders to evaluate their claims.

California targets addictive social-media design

California has banned several social-media features for users under 16, including infinite scroll and autoplay, according to the roundup. The law is described as the first of its kind in the United States.

These features remove natural stopping points. Infinite scroll eliminates the need to choose a new page, while autoplay begins the next piece of content without requiring an active decision. Restricting them for younger users reflects a policy judgment that product design can affect behavior and that children may warrant additional protection.

The law also introduces rules involving AI and companion chatbots. The source material does not provide the full text or implementation details, so its practical effect will depend on how the provisions are interpreted and enforced.

For parents and young users, the immediate takeaway is narrower than a general ban on social media: the measures target particular engagement mechanisms and AI-related services. They do not, based on the information provided, eliminate all access to social platforms for people under 16.

Departures and disagreements inside the AI-safety field

Two AI researchers reportedly left Anthropic and Google over concerns about safety risks, following Jacob Coxon’s widely discussed departure from Anthropic. The departures have become part of a wider argument over whether leading AI companies are moving quickly enough to manage the systems they develop.

Elon Musk characterized the concerns as a “setup” and a “psyop,” while the roundup reports that fears about AI are contributing to calls in Congress for stronger regulation.

Personnel departures are not, by themselves, evidence that a company’s safety practices have failed. They can, however, reveal disagreement about priorities, risk thresholds, governance, or the pace of deployment. The significance of each departure depends on the individuals’ stated reasons and on whether their concerns are supported by evidence about specific systems or decisions.

Critical infrastructure becomes a target for AI defenses

Sam Altman has been pitching OpenAI’s cybersecurity capabilities to power companies, following reports of AI-assisted attacks on critical systems. The meetings suggest that energy infrastructure is becoming a prominent arena for both AI-enabled threats and proposed AI defenses.

Power networks are attractive targets because disruption can affect large numbers of people and businesses. Security teams must also protect operational technology alongside modern computer systems.

The source material does not establish how effective OpenAI’s defenses are or whether any utility has adopted them. For now, the development is best understood as a sign of growing commercial interest in using AI to defend essential infrastructure, not as proof that AI can reliably secure the power grid.

Altman also reportedly told OpenAI employees that the company is open to slowing down AI. That position reflects a continuing tension in the industry: companies want to develop and deploy capable systems, while safety concerns may require delays, restrictions, or additional testing.

Music companies begin experimenting with AI

After years of conflict between the music industry and AI developers, Universal is partnering with ElevenLabs on an AI remix platform, according to the roundup.

The shift does not resolve the central disputes over copyright, consent, compensation, or the use of artists’ voices and recordings. It does suggest that parts of the industry are moving from opposition alone toward controlled experimentation.

An AI remix platform could create new tools for artists and listeners, but its legitimacy will depend on the terms governing the underlying music. Questions about who authorizes a remix, who receives payment, and whether an artist’s identity can be reproduced remain central to any commercial model.

Sources

• The Download: biotech’s future and cheaper, cleaner steel, MIT Technology Review, September 11, 2026.