Exploring emerging biosecurity threats from AI-designed proteins and the scientific responses to protect our biological future
Imagine a world where deadly toxins could be designed not in a high-security laboratory, but by artificial intelligence responding to simple text prompts. Where the building blocks of life can be ordered online and assembled with increasingly accessible tools. This isn't science fiction—it's the emerging reality of biotechnology that is forcing a revolutionary rethink of how we protect humanity from biological threats.
The same technologies that promise revolutionary medicines, sustainable fuels, and agricultural breakthroughs also carry unprecedented potential for misuse.
As one recent study starkly warns, our current biosecurity safeguards are being outpaced by AI-powered protein design tools that can create entirely novel biological sequences with potentially dangerous functions 1 . The very tools accelerating our ability to heal may also lower barriers to causing harm. This article explores how scientists, policymakers, and security experts are racing to close this gap—building what many call an "invisible shield" to protect our biological future.
Revolutionary treatments for diseases using gene editing and synthetic biology
Potential misuse of biotechnology to create novel pathogens and toxins
For decades, biosecurity screening has relied on a straightforward principle: compare new genetic sequences against databases of known threats. Much like security personnel checking passports against a watchlist, this sequence-based screening looks for matches to pathogens and toxins that have been identified and cataloged. This approach has served us reasonably well for natural pathogens and historically known biological threats 1 .
Compares DNA sequences against databases of known pathogens and toxins.
Effective against known threats but misses novel designs
Generative AI creates novel protein sequences with little similarity to known threats.
Traditional screening misses most AI-designed threats
The game-changer is artificial intelligence. Generative protein design tools can now create completely novel protein sequences with specific functions—including harmful ones—that bear little resemblance to anything found in nature. As Professor Natalio Krasnogor of Newcastle University explains, these AI systems pose a "growing biosecurity risk because they have the potential to produce functionally dangerous proteins with little homology to sequences of concern" 1 .
This creates what security experts call a biosecurity blind spot. A toxin that looks entirely new to screening algorithms could potentially be ordered as synthetic DNA and slip through existing security checks. The technical barriers to turning these digital designs into actual biological agents remain significant, but they're steadily decreasing as laboratory capabilities advance worldwide 1 .
In response to this emerging threat, a broad collaboration across industry, academia, and government laboratories has proposed a revolutionary approach. Published in Science in October 2025, research led by Bruce J. Wittmann and Eric Horvitz of Microsoft's Office of the Chief Scientific Officer demonstrates what they call a "function-based screening" method 1 .
Rather than simply looking for sequence similarity to known threats, their hybrid approach integrates functional prediction algorithms that can flag synthetic genes encoding hazardous functions—such as enzymatic activity linked to toxins—even when their sequence signatures appear novel 1 . This represents a substantial shift from asking "Does this sequence match something dangerous?" to "Could this sequence produce something dangerous?"
Professor Francesco Aprile of Imperial College London praises this approach as "a practical, timely safeguard" that enhances current DNA synthesis screening and "establishes a solid foundation for continued optimisation" 1 .
The researchers tested their function-based screening method through a systematic comparison with traditional approaches:
| Screening Method | Basis for Detection | Strengths | Limitations |
|---|---|---|---|
| Sequence-Based (Current Standard) | Genetic similarity to known threats | Effective against known pathogens; Well-established | Misses novel or engineered sequences |
| Function-Based (New Approach) | Predicted biological activity | Can identify novel threats with dangerous functions | Computationally intensive; Requires advanced algorithms |
| Hybrid Approach (Proposed) | Combines sequence and function analysis | Comprehensive protection; Future-proof | More complex to implement |
The methodology followed a clear, step-by-step process:
Researchers first trained machine learning models to recognize sequence patterns associated with specific biological functions, not just similarity to known dangerous sequences.
The team then used AI protein design tools to create novel sequences with potential toxic functions that wouldn't be flagged by conventional screening.
They implemented a system that ran both traditional sequence matching AND functional prediction algorithms on each sequence.
Finally, they validated their results by comparing which dangerous sequences would be caught by each method alone versus the hybrid approach.
The results demonstrated a dramatic improvement in identifying novel threats. While traditional screening caught only about 12% of AI-designed toxic sequences, the function-based approach identified 96% of these novel threats 1 . This hybrid method could prevent potentially dangerous synthetic genes from being synthesized while minimizing false positives that might hinder legitimate research.
While AI-designed proteins represent a cutting-edge challenge, the biosecurity threat landscape is expanding across multiple fronts. A new framework from the RAND Corporation assesses how technology maturity and diffusion may lower barriers to biological weapon development 6 .
Benefits: Gene therapy, agricultural improvement
Risks: Genetic modification of pathogens
Benefits: Drug discovery, protein design
Risks: Creation of novel pathogenic agents
Benefits: Rapid vaccine development, research
Risks: Production of custom pathogen sequences
The RAND study modeled three concerning adversary scenarios enabled by these technologies: genetic modification of existing pathogens, resurrection of previously existing or difficult-to-acquire pathogens, and the creation of entirely novel pathogens 6 . Experts concluded that several emerging technologies—particularly CRISPR and foundational AI models—are likely to mature within the next decade, thereby potentially lowering barriers for hostile actors 6 .
Not all technologies advance at the same pace, however. Some tools, such as cloud labs and benchtop DNA synthesizers, weren't expected to fully mature in the same time frame, suggesting that risks are uneven across technologies 6 . The convergence and combination of multiple tools—particularly when coupled with growing accessibility—will shape the biosecurity concerns of the coming decade.
Nations and international organizations are recognizing the urgent need to adapt to these evolving threats. The European Union has identified biosecurity as an increasing threat in our shifting geopolitical climate, noting that "advances in biotechnology, Artificial Intelligence (AI), and the capacity to engineer biological components of life, increase possibilities for scientific progress and innovation" while also bringing "heightened risks of the release of biological agents either by accident or through weaponisation" 2 .
Current voluntary frameworks like the International Gene Synthesis Consortium screening guidelines may require revision to incorporate function-based detection, ensuring no jurisdiction becomes a "screening haven" 1 .
The EU currently lacks regulation to oversee DNA synthesis screening, creating potential gaps in monitoring access to synthetic genetic materials 2 .
Current EU guidelines for dual-use research of concern were written in 2019 and need updating to reflect current challenges and risks in the biotechnology landscape, especially those advanced by AI 2 .
Similar to how satellite surveillance monitors nuclear threats, significant investment is needed in a new domain of biological intelligence to detect, deter, and neutralize biological threats 7 .
"Biosecurity requires U.S. leadership in emerging biotechnology. Leadership means innovation. We must advance the field to secure it" 3 .
Modern biosecurity research relies on sophisticated tools spanning biological, computational, and analytical domains. Here are essential components of the biosecurity research toolkit:
The challenges in biosecurity are profound, but not insurmountable. As technologies continue to evolve, so must our approaches to safeguarding their use. The transition from sequence-based to function-based screening represents exactly the kind of innovative thinking needed to address emerging risks.
What makes this moment particularly critical is the convergence of multiple technological trends—from the democratization of biotechnology to the explosive growth of artificial intelligence. As the RAND Corporation's research suggests, no single technology uniquely drives risk, but the combination and convergence of tools creates new vulnerabilities 6 .
The path forward requires a balanced approach that neither stifles innovation nor ignores real dangers. It demands international cooperation, sustained investment in defensive technologies, and what experts call a "culture of responsibility" across all research stages and stakeholders 2 . In the words of one report, "We can make a world in which we are safe from preventable biological harms. We can make a world in which COVID-19 is the last pandemic humanity ever faces" 3 .
The invisible shield of biosecurity may never be complete, but through continued scientific ingenuity and global collaboration, it can become strong enough to protect both our biological future and our innovative spirit.