The strongest evidence-based answer is about bottlenecks, not about impossibility. A 2025 consensus report from the National Academies of Sciences, Engineering, and Medicine concluded that no available AI-enabled biological tool can design a virus from scratch. Designing a pandemic-capable pathogen would need biological knowledge and data that don’t yet exist, and it would still require physical building and testing that software can’t skip.
That is a claim about today’s tools. It isn’t a claim that AI biosecurity risk is zero, and it isn’t a probability of extinction. This article explains what the assessment found, where its limits are, and what could change the picture.
As an Amazon Associate I earn from qualifying purchases.
What the National Academies actually concluded
The report is The Age of AI in the Life Sciences: Benefits and Biosecurity Considerations, a 2025 consensus study from the National Academies of Sciences, Engineering, and Medicine. Its central finding on this question is stated in Chapter 3:
Recommended Free Tools
“The relative lack of biological and mechanistic understanding about virulent phenotypes and the paucity of high-fidelity biological data mean that AI-enabled biological tools currently cannot be used to de novo design and subsequently build complex biological systems that can successfully replicate as transmissible biological agents with epidemic or pandemic potential.”
#1 Best Overall
Sale
3M OV/P100 Odor/Particulate Multipurpose Respirator 65021, Medium | Reusable Respirator, 1-Facepiece and 1-pair of Replaceable NIOSH-Approved 60921 Combination Cartridge/Filters
- NIOSH-APPROVED: Organic Vapor/P100 combination cartridge helps reduce expose to certain organic vapors and provides at least 99.97% filtration efficiency against solids, liquid aerosols, and oil-based particulate
- OIL-PROOF FILTERS: Included P100 particulate filters are oil-proof (Use according to manufacturer’s time use restrictions when oil aerosols are present)
- USE FOR: Relief from certain non-harmful particles and annoying odors
- KIT INCLUDES: One facepiece and one pair of replaceable organic vapor/P100 combination cartridges (#60921)
- 3M BAYONET CONNECTION: Makes it compatible with a wide range of 3M cartridges and filters
That sentence is the committee’s conclusion, not a paraphrase. Two phrases carry most of the weight. “Currently” ties the finding to the tools reviewed. “De novo design and subsequently build” covers the whole pathway, from a digital design to a working, transmissible agent.
Why a model’s output isn’t a working organism
The popular picture is an AI that invents a supervirus and a lab that instantly makes it. The report points to several places where that picture breaks down.
Rank #2
- Our Premium Respirator Kit offers unparalleled protection for your face and eyes in a wide range of hazardous environments. This complete suit includes a half face shield, safety glasses, and an ample supply of replaceable filters, making it the ideal choice for professionals and DIY enthusiasts alike.
- The half face shield and safety glasses provide a secure barrier against harmful substances, including gases, paints, fumes, and more.
- Superior Filtration: Equipped with a dual filtration system, the respirator effectively blocks 97% of airborne particles.
- Versatility: Ideal for a variety of applications, including gas, painting, machine polishing, organic vapors, staining, car spraying, sanding, cutting, and general DIY work.
- Complete Kit: Includes one half face shield, one safety glasses, ten cotton filters, two filter cartridges, and two filter covers for comprehensive protection.
Biological data is limited and noisy
AI tools are only as good as the data they learn from. The committee describes a shortage of high-fidelity biological data, particularly data linking genetic sequence to the traits that make a virus dangerous.
Complex traits are hard to predict
How a virus behaves depends on intricate interactions with a host and with its environment. The report notes that the phenotype information needed to capture this isn’t adequately represented in existing datasets. A model can propose something that looks plausible on a screen. That is very different from showing that the thing will replicate, spread, or cause a particular outcome.
Rank #3
- PROTECTION : The AirGearPro G-500 Respirator mask and Safety Goggles are CE certified to ensure your protection during painting, woodworking, construction, sanding, renovation, demolition etc.
- DUAL FILTRATION SYSTEM (A1P2): Our respirator is designed with a dual filtration system of class A1P2 to guarantee your protection against vapours and particles. The mask cartridges filter out organic vapour with boiling points above 65°C. The P2 dust filter filters at least 98% of airborne particles.
- REUSABLE MASK: Our dust mask makes it possible to replace the filters when they reach the end of their life and lets you continue to use the mask with optimal filtration, thus reducing the impact on the environment.
- LIGHTWEIGHT & COMFORTABLE: Thanks to a lightweight structure, adjustable straps and the use of high-quality materials that feel soft on the skin, this respirator mask is ideal for long work sessions, as it will not tire you.
- EASY TO ASSEMBLE AND USE: User's manual in English with clear instructions and safety precautions. If you any questions, feel free to contact our customer service.
Design is only one stage of a longer cycle
Biological engineering is usually described as a design-build-test-learn loop. Models can give data-driven guidance for the design step. The report describes a gap between digital design and physical construction and testing, and physical designs still have to be built and tested. Each step can fail, and an AI proposal has to survive all of them.
This article deliberately stays at that high level. How pathogens are constructed isn’t something to spell out, and the reader-relevant point is simple: protein-design and property-prediction tools existing doesn’t mean they can produce a viable, transmissible virus.
Rank #4
- NIOSH-APPROVED: 3M Particulate Filter 2097, P100 provides at least 99.97% filtration efficiency against solids and liquid aerosols, as well as oil-based particulates
- SPECIAL CARBON LAYER: Features a carbon layer which helps reduce the odor of nuisance organic vapors* including certain solvents, degreasers, and resins
- USE FOR: Professional removal of mold, lead-based paints, woodworking and working with fiberglass insulation and to help reduce nuisance odors
- FOR WORKPLACE/OCCUPATIONAL USE ONLY
- KIT INCLUDES: One facepiece and two pair of replaceable P100 filters (#2097)
What current AI tools can do
The assessment doesn’t call AI irrelevant to biosecurity. Its findings are narrower and more specific.
Free tools Windows power users keep installed
One-click scans. No signup required.
- Simple biomolecules: Available tools can design or redesign some simple biomolecules, including toxins. The report characterizes the likely scale of threat from these as local rather than epidemic or pandemic.
- Features of existing pathogens: Some tools may model particular features linked to virulence in pathogens that already exist. Performance is constrained by insufficient datasets and by the complexity of biological systems.
- Misuse concerns: The committee considered harmful applications and recommended monitoring and mitigation alongside support for beneficial life-sciences research, according to the National Academies’ news release on the report.
So the finding isn’t that nothing can go wrong. It is that the capabilities found are limited in scale and don’t amount to an end-to-end route to a pandemic agent.
Best Value
- High-Efficiency Protection: TICONN safety gas mask features a dual filtration system that effectively blocks vapors, gases, smoke, pollen, and dust; A front-mounted one-way exhalation valve helps release exhaled air for easier breathing
- Safe and Comfortable: This respirator mask with filters is CE certified for quality and safety; It features a food-grade elastic silicone seal and adjustable straps for a secure, comfortable fit on most head shapes
- Used with Safety Glasses: Equipped with anti-fog safety goggles, the paint mask offers reliable eye protection against debris, splashes, and dust; The anti-fog coating ensures clear vision even during extended use
- Wide Applications: Easy to assemble and use, this paint respirator mask is suitable for scientific work, outdoor operations, welding, railways, manufacturing, wood sawing, workshops, eye protection DIY, laboratories, chainsaw workers, construction, deserts, electricians
- TICONN Quality Commitment: TICONN stands behind the quality of our products; If for any reason you are not satisfied with your purchase, please contact us
How the stages compare
| Axis | Lower concern (per the 2025 assessment) | Higher concern |
|---|---|---|
| Capability stage | Digital design, where tools give some guidance | Physical build and test, where the report finds a gap that digital design alone doesn’t close |
| Biological complexity | Simple biomolecules such as toxins, which tools can design or redesign | Complex, replicating, transmissible agents, which the report says tools can’t currently design and build |
| Scale of plausible harm | Local, for toxin-type threats | Epidemic or pandemic, which the report ties to capabilities not currently available |
| Time horizon | Tools assessed in 2025 | Future developments listed below |
What could change the assessment
The report names developments that would matter. These are watchpoints, not capabilities the committee says exist now:
- More accurate prediction of pathogenesis and transmissibility.
- AI design of fully replicating infectious agents.
- Designs that need little or no experimental validation.
- Advances in AI-driven, automated laboratories.
The report describes de novo design of a pandemic-capable virus as a major potential uplift in capability. Because the conclusion rests on the tools and evidence reviewed, it can be overtaken by progress. The report is from 2025, and anyone reading it later should check whether newer assessments have revised its findings on any of these four points.
What this evidence doesn’t tell you
Read the finding for what it is. The report is a technical capability assessment of AI-enabled biological tools and the state of the science. It says explicitly that it isn’t a threat assessment. It also notes that empirical research directly measuring the biosecurity risks of these tools is limited.
Two things follow. First, the sources used here give no quantitative estimate of the probability that AI-enabled bioweapons could cause human extinction, and no number should be read into them. Second, “isn’t likely” is a reasoned judgment about bottlenecks that exist today, not a guarantee. The defensible position is between the two extremes: today’s tools don’t provide a route to a novel pandemic virus, and monitoring the watchpoints above is still warranted.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
