Artificial intelligence combined with biological data and manufacturing capabilities creates a powerful dual-use technology that can accelerate both lifesaving treatments and dangerous pathogens. The same AI tools that help researchers discover new drugs in weeks instead of years could enable a person with minimal training to design a biological weapon. This convergence, known as AIxBio, represents one of the most consequential governance challenges facing the United States, according to a new analysis from Harvard's Belfer Center for Science and International Affairs.

What Makes AI-Powered Biology So Dangerous?

The threat isn't hypothetical. Large language models (LLMs) and open-source biological tools can now substitute for decades of specialized knowledge that once protected against misuse. Researchers can search scientific literature, troubleshoot experimental workflows, and learn complex protocols through AI assistants in ways that were previously impossible without years of hands-on training. This "barrier reduction" is the most immediate risk: expertise that once required a PhD and a well-equipped laboratory is becoming accessible to anyone with internet access and determination.

The long-term danger is even more severe. Open viral sequence databases, genome language models like Genos, CRISPR tools, and nucleic acid synthesis technology could theoretically enable malicious actors to enhance pathogens for increased transmissibility, virulence, host range, or immune evasion. Viral design is particularly concerning because viral genomes are shorter and easier to synthesize than more complex biological systems.

Where Is U.S. Policy Falling Short?

Current safeguards are fragmented and largely voluntary. Frontier AI model safety measures depend on corporate self-regulation rather than mandatory standards. DNA synthesis screening, which should be a critical access control, does not bind the full private market. Many synthesis providers lack customer verification, sequence screening using best practices, or systems to detect fragmented orders that could circumvent safety checks.

Outbreak detection remains reactive rather than proactive. Public health systems often identify biological threats only after spread has already begun, a timeline that becomes catastrophic if the threat is engineered rather than natural. The COVID-19 pandemic demonstrated this vulnerability; despite advance warnings about coronavirus risks, the world was unprepared when SARS-CoV-2 emerged.

How Can Policymakers Close These Gaps?

• AI Model Safeguards: Establish a government-authorized private regulatory market where licensed technical auditors enforce AI-biosecurity standards for frontier models, using NIST or similar technical standards to augment private audits.
• Synthesis Screening: Require universal screening of synthetic nucleic acid orders longer than 50 nucleotides across the private sector, with mandatory customer verification, sequence screening, and reporting of failed legitimacy checks.
• Early Detection Systems: Build an AI-enabled epidemic intelligence system that integrates wastewater monitoring, bioaerosol sampling, genomic surveillance, and metagenomic next-generation sequencing (mNGS) to detect anomalies before they spread.
• Equipment Licensing: Create a federal licensing regime for sensitive laboratory equipment, requiring buyer verification, pre-purchase approval, ownership registration, and transfer tracking.
• Vaccine Manufacturing: Secure scalable mRNA vaccine manufacturing capacity through government-backed private partnerships to enable rapid response to novel biological threats.

The Belfer Center report emphasizes a critical tension: overreaching regulation could slow U.S. biomedical and AI innovation, while weak regulation could permit catastrophic misuse. The policy challenge is targeting vulnerabilities without sacrificing the technological dominance that strengthens national security.

What About the Civilian Benefits of AI Biology?

The upside of AIxBio is substantial and shouldn't be abandoned in pursuit of safety. The same technology stack enables faster small-molecule drug discovery, more efficient messenger RNA (mRNA) design, improved CRISPR therapies, individualized cell therapies, and breakthroughs in personalized medicine. Meanwhile, researchers are already demonstrating real-world progress. Chinese researchers recently developed OneGenome, an open-source genomic AI system that combines a human-genome foundation model named Genos with large language model capabilities to assist clinicians in diagnosing rare diseases.

OneGenome links DNA variants directly with clinical literature and phenotypic information, functioning as an advanced decision-support tool for precision medicine. The platform surpassed conventional gene models and general-purpose AI tools across several diagnostic and medication-guidance benchmarks. A related platform called DeepRare, developed by researchers at Shanghai Jiao Tong University and Xinhua Hospital, achieved a 57.18% top-ranked diagnostic rate using phenotype data alone, rising above 70% when whole-genome or exome information was incorporated.

These advances matter because rare disease diagnosis remains a critical bottleneck in healthcare. A 2024 study published in Nature Genetics reported that genome sequencing provided plausible diagnostic answers for 29.3% of participants with suspected rare conditions. A 2025 study of 1,452 Korean families demonstrated a 46.2% molecular diagnostic yield, which altered clinical management in 18.5% of diagnosed cases.

Why Does This Matter Now?

AI development is accelerating rapidly, and strategic competition with China makes American technological dominance essential. The United States cannot assume that rivals will accept the same ethical limits, nor can it sacrifice technological advantage in the name of caution. Yet the window for establishing governance frameworks is closing. The steps taken today will determine whether AI-biosecurity becomes manageable or catastrophic.

The challenge is not choosing between innovation and safety. It's building governance structures that enable both. That requires moving beyond voluntary corporate safeguards to mandatory standards, closing gaps in DNA synthesis screening, and investing in early detection systems that can identify threats before they spread. Without these changes, the same technology that promises to cure rare diseases and accelerate drug discovery could enable someone with minimal training to engineer a biological catastrophe.