The Double-Edged Helix: How AI Is Reshaping Biosecurity Risks and Defenses

The same AI tools that enable scientists to design novel proteins and accelerate drug discovery are now being scrutinized for a darker potential: helping malicious actors engineer transmissible biological threats at epidemic scale. In response, the National Academies of Sciences, Engineering, and Medicine has launched a consensus study under Executive Order 14110 to assess how artificial intelligence may raise or reduce biosecurity risks—and to recommend concrete mitigations. The study, which comes amid a flurry of warnings from AI executives and a notable absence of comprehensive federal regulation, marks the most authoritative U.S. government-backed effort yet to grapple with the dual-use nature of AI in biology.

The National Academies committee is tasked with evaluating three core questions: how AI can increase biosecurity risks—including generative AI trained on biological data; the national security implications of pathogen and omics datasets used to train AI; and how AI can reduce biosecurity risks, for instance through better coordination of data and compute resources. The study was mandated by the Biden administration’s AI executive order, which recognized that AI’s ability to design, synthesize, and troubleshoot biological agents could expand access to capabilities once limited to highly trained specialists.

This concern is not hypothetical. In a 2024 interview, Dario Amodei, CEO of Anthropic, warned that AI could “greatly widen the range of actors” capable of launching a large-scale biological attack. Sam Altman, CEO of OpenAI, has similarly called for regulation of AI models that could help create novel biological agents. Their companies have begun to act: in June 2026, Anthropic released a proposed policy framework addressing biological weapons, cyber risks, loss of control, and automated R&D dangers—the most recent concrete development in the source set.

Yet the policy landscape remains fragmented. A Congressional Research Service (CRS) report notes that there is no single overarching U.S. federal biosafety or biosecurity law with enforceable penalties beyond the Federal Select Agent Program, and no broad federal AI law establishing regulatory authority for AI in biology has been enacted. That means the National Academies study is operating in a vacuum of statutory guidance, relying on expert consensus to shape future rules.

The stakes are high because the pathways for misuse are well-documented. Researchers have identified that AI can help design novel toxins, viral pathogens, and altered pathogen proteins. It can also assist malicious actors in evading nucleic-acid synthesis screening and other safeguards that currently prevent dangerous DNA sequences from being ordered online. At the same time, AI offers powerful tools for biosurveillance and countermeasure development, creating a classic dual-use dilemma rather than a one-sided threat.

A 2026 Nature report captured the active expert debate about how worried the field should be, noting particular concern about AI-designed pandemic viruses. The fear is that AI could lower the technical barriers so dramatically that a lone actor with access to a large language model and a DNA synthesizer could recreate or modify a known pathogen—or even design something entirely new.

Historically, biological weapons development required deep expertise in virology, molecular biology, and wet-lab techniques. AI changes that calculus by acting as a “co-pilot” for design, troubleshooting, and optimization. The CRS report and the National Academies study both highlight that the risk is not just about existing pathogens but about the potential for AI to enable the creation of threats that evade current countermeasures.

For biotech firms, pharmaceutical companies, and AI labs, the practical implications are immediate. The push is toward stronger model safeguards, such as restricting the ability of AI systems to output detailed protocols for pathogen engineering or to help users circumvent DNA synthesis screening. Companies that provide DNA synthesis services are also under pressure to implement better screening of orders against known threat sequences and to verify the identity and intent of customers.

Tighter oversight of high-risk biological data is another likely outcome. The CRS report notes that large biological datasets used to train AI models—such as protein structure databases or viral genome repositories—could be double-edged swords. While they accelerate research, they also provide a rich resource for anyone seeking to design harmful agents. Expect future regulation to require data access controls, usage logging, and possibly licensing for certain types of biological training data.

For investors and startup founders, the message is clear: biosecurity is becoming a regulatory and reputational risk factor. Companies that proactively adopt screening and safety measures may gain a competitive advantage, while those that ignore the issue could face backlash or legal liability. The Anthropic framework, though voluntary, sets a benchmark that others may be measured against.

The National Academies study is expected to produce recommendations within the next year. Those recommendations could form the basis for legislation or executive action, especially if Congress decides to fill the regulatory gap. Meanwhile, the Anthropic framework from June 2026 will likely be refined and possibly adopted by other AI labs. The key question is whether voluntary measures will be enough to prevent a real-world biosecurity incident—or whether it will take a crisis to catalyze comprehensive regulation. For now, the clock is ticking, and the dual-use helix of AI and biology is spinning faster than the policy machine can keep up.