Executive Summary
If we fail to act, an extinction-level event* within the next 100 years is probable. Synthetic biology is advancing faster than our ability to secure it, and the same tools that promise medical revolutions also lower the barrier to weaponizing biology.
Bioengineered pathogens are coming. OpenAI and Anthropic both recognize the threat and have Red Teams assigned to the problem. Anthropic recently published that Claude advanced from performing below world-class virology experts on a benchwork evaluation to comfortably surpassing them within a year. Despite this information being public, the threat remains outside the broader public discourse, and efforts to improve biodefense are sparse and underfunded.
I still believe we won’t fail to act, and am highly optimistic about a safe and abundant future. To reach this future, we must architect a proactive defense network capable of responding to escalating biowarfare and bioterrorism: a massive, adaptive biosecurity stack that spans detection, prediction, manufacturing, vaccination, active neutralization, and cyber-biological security.
This document outlines seven categories of technologies and infrastructure that must exist to establish global biowarfare and bioterrorism resilience, each with major technical gaps, ethical dilemmas, and differing levels of commercial viability. This is not comprehensive, but it does cover the main categories. It concludes with a list of critical questions that need to be answered to build out those categories of technologies and infrastructure.
*I define extinction-level here as profound population loss or partial collapse of global civilization and hope this is hyperbole.
1. Biointelligence (Detection & Prediction)
Goal: Detect and characterize new pathogens as they spread.
Key Functions:
Detect emergence: Pathogen-agnostic surveillance such as wastewater and agricultural detection systems that transmit data instantly to centralized models
Predict threat: ML models that infer virulence and transmissibility across species
Prioritize: Identify which pathogens are dangerous enough to develop countermeasures
What Needs to Exist:
Low-cost pathogen-agnostic detection methods: Closer to the cost of MALDI-TOF mass spec (<$10/sample) but without requiring organisms to be culturable or already in the reference set
Distributed detection: Detection systems distributed globally at key points
AI-enabled pathogen modeling: Combining epidemiology, genetics, and immunology to forecast threat level in real time by predicting transmissibility and virulence
Massive data collection, integration, and analysis: Integrating hospital data, mobility trends, and genomic surveillance to detect anomalies before symptoms spike + government surveillance
Technical Gaps:
Cheap, scalable detection: Sequencing is expensive for continuous deployment across thousands of sites
Intelligence: Models currently fail to robustly predict danger from new pathogens
Speed: In a worst-case scenario with high R₀ and long incubation, we miss the window by a wide margin
Commercial Pathways:
B2B companies selling early-warning signals to agriculture, pharma, insurance, or hedge funds
B2G public-private model, powering government surveillance networks
2. Biomanufacturing (Distributed Infrastructure)
Goal: Compress time from design to global delivery for biowarfare countermeasures.
Current Paradigm:
It takes months to scale an already-validated countermeasure
All major systems are centralized, expensive, and require extensive manual labor
Cell-free platforms exist but are too costly and currently volume-limited
Manufacturing is consistently moving out of the US
What Needs to Exist:
US-based pipelines: We currently rely on China and other countries for most of our manufacturing. This is unsustainable in a world where biowarfare is a reality.
A Department of Commerce study reportedly found that 97% of all antibiotics in the United States come from China
80% of all the active pharmaceutical ingredients (APIs) used to make drugs in the United States are produced abroad
Potential manufacturing technology improvements
Sequence-to-dose pipelines that can turn a sequence into material countermeasures in <48 hours
Self-amplifying RNA
Commercially-viable GMP-in-a-box
Cell-free synthesis systems that can produce RNA, proteins, or biologics anywhere in portable and autonomous manufacturing units
Automated QC & analytics, replacing reliance on trained technicians and getting closer to eliminating the process=product regulatory issue
Regulatory, compliance, and legal structure changes
Licensing platforms for rapid-response CDMOs
Automated compliance
Commercial Pathways:
Manufacturing
Advent of AI also means more biologics produced every year, meaning manufacturing demand will go up anyways, so there is an opportunity now to build dual biodefense and pharmaceutical manufacturing infrastructure
SwiftScale failed, so there are clearly some limits as to what can be built out in terms of cell-free –– likely they failed for several reasons, including:
Too early
Tried to do glycosylation
Cell-free in large batches is still prohibitively expensive
Regulatory, Compliance, and Legal Structure
Better licensing platforms could be commercially viable now
Automated compliance could be commercially viable now
3. Biodefense (Immunity)
Goal: Engineer immunity as quickly as possible.
What Needs to Exist:
mRNA vaccine libraries: Easily swappable payloads for different known threats
Broad-spectrum vaccines & AI-optimized immunogen design: Broader protection against future variants and engineered mutations, towards more universal vaccines
Novel distribution mechanisms: Take advantage of modern shipping systems by distributing the vaccines to a person’s door via patches, creams, etc.
Prophylactic antibody-encoding mRNA: get around immune system failures and produce the antibody directly
Commercial Pathways:
B2G public-private model
Wildlife disease control/preventing zoonoses
4. Bio-Offense (Programmable Antivirals)
Goal: Actively destroy engineered bioweapons.
What Needs to Exist:
Biologics that only activate in infected cells or in presence of specific markers (e.g. the virus it is targeting) and are off-the-shelf programmable
CRISPR-based antivirals, with guide RNAs targeting only known pathogen signatures
mAbs, potentially self-stable, could reuse Fc backbone and swap out Fabs, could also encode them in mRNA
Rapid deployment:
Formats that are ideally heat-stable, shelf-ready, etc.
Potential for transmissible antivirals to out-compete the virus by spreading faster and circumvent the limitations of manufacturing (high dual-use risk)
Key Technical Needs:
Safety: Limit off-target effects
Speed: Design-to-dose as quickly as possible
Obfuscation: Prevent hostile re-use of code
Commercial Pathways:
B2G public-private model
5. Containment and Environmental Defense
Goal: Build a pathogen-agnostic, physical layer of defense: secure labs, PPE, and clean-environment infrastructure that prevent pathogen spread.
What Needs to Exist:
Secure Labs
Automated containment systems: Minimize human error with robotic handling (this has high dual-use risk) and include self-sterilizing surfaces, HEPA/hydrogen peroxide, bubble isolation
Next-Gen PPE & PPE distribution
Cheap PPE storage: Mass produce n95-n99 respirators and store units globally for instant distribution
Better PPE: Respirators that last longer, are cheaper to manufacture, and are more effective
Environmental Biosafety Tech
Commercial Pathways: (Need regulatory tailwinds)
Sell into BSL2, 3 and 4 labs
B2G stockpile creation
Agriculture or food safety
6. Cyber-Biosecurity
Goal: Secure the digital and molecular layer of biology to prevent theft, misuse, and sabotage.
What Needs to Exist:
Regulatory Structure:
Policy in place, preferably global, for how to ensure generative AI is not misused. Limit people’s ability to download open-source models, though this will only slow progress temporarily.
Policy in place, preferably global, preventing mail-order sequences from being used for bioweapons. (It currently costs less than $100,000 to synthesize a smallpox-like virus via mail-order sequences.) This also requires biointelligence software that can detect when something novel is dangerous. Current systems can do this sometimes, but, with no global requirements to have them in place, many orders never run through those security checks. And the current systems are not good enough at catching novel pathogens. (Need the tools described in (1).)
Preventing LLMs from generating viable pathogen designs:
Biointelligence: Would again need the functional understanding described in (1) to detect potentially problematic sequences
Biofirewalls: Sandbox everything as meaningfully as possible with detection built in –– would probably have to be legislated, as not all AI companies are going to comply by choice
KYC and better classifiers for detecting dangerous requests
Protecting Data: Systems need to be in place to protect hidden barcodes from foreign nations entering our data pools that could allow them to later deconvolve our systems
Securing dual-use molecular designs: Can you encrypt nucleic acids to conceal function or resist sequencing? Can you decrypt them if so?
Deadman switches: Sequences degrade after x replication cycles without host
Structure-only function: Sequences where function is only obvious when tertiary folding occurs (or other folding e.g. with a substrate that is impossible to predict through something like DeepMind)
Masking: Code disguised as host or contains numerous meaningless additions such that our best software can’t detect the real GoF changes
Non-canonical base or codon incorporation: You would have to include nonstandard tRNA with delivery…or code for it, which seems unlikely to be possible
7. Regulatory
Goal: Compress approval timelines as much as possible.
Regulatory is by far the biggest cause of timeline elongation when it comes to getting vaccines and other MCMs to the population.
The design of the COVID mRNA vaccines took 2 days, but it took almost a year to approve them. EUA’s during COVID took 3-4 weeks to approve from submission.
What Needs to Exist:
End-to-end platforms for vaccines and other MCMs that are partially or fully pre-approved (e.g. swap-in, swap-out technology with pre-approved backbone and allowance or pre-approval for various “swap-ins”)
Broader spectrum vaccines and other MCMs
Threat-tiered evidence standards that default to EUA pathway/Animal Rule-like skips
Questions
Scope
How fast can the most well-designed biowarfare weapon spread?
What is the best design for a bioweapon? i.e. what do you optimize for and how would you counteract that?
Who is likely to want to design something like this? Who are the first targets?
Commercialization
Key: Who pays for countermeasures before an outbreak?
How big will the biosecurity market be in the next ~5 years and where will the money be coming from?
What are the drivers changing the market? e.g. federal funding, technological advancements, population awareness, and government initiatives aimed at enhancing the preparedness against Chemical, Biological, Radiological, and Nuclear (CBRN) threats.
Capital landscape: What is the sum total of money publicly directed towards biosecurity and where is it coming from? For what is it primarily being used? How much of it is going towards true cutting edge work preparing for a manmade pandemic?
Between large incumbents and startups, the landscape of all the current biosecurity companies: What do they sell and to whom?
How much would drug developers or financial firms pay for early-warning data? Can early access to virulence data create enough alpha for anyone, especially besides agriculture?
Can private insurers or reinsurance firms become buyers in a pre-pandemic market?
Regulatory, Ethics, and Governance
Regulatory and governance landscape: What does the landscape look like for all regulatory bodies involved with biosecurity, with a focus on the US? Who monitors current abuses and who should monitor them? e.g. Why do only 4 people work at the BWC?
If we know how to solve the biggest bottlenecks to faster approval times, how do we then implement those changes? Who is responsible for this?
Should there be treaty-backed limits on dual-use sequence design?
What would a version of the BWC with actual oversight look like and how could we make it happen?
Technical
Key: What new infrastructure/technology is needed to compress pathogen release-to-dose from months to hours?
Are there any technologies too good to ignore? That is, what new biodefense-relevant technology is a massive-fold improvement on existing tech? e.g. cell-free speed of manufacturing
Which methods perform pathogen-agnostic testing of a sample and are 10-100x cheaper than metagenomic sequencing?
How much cheaper could you make metagenomic sequencing without making a better sequencer?
How do you build better classifiers and KYC from the model side?
How hard is it to predict pathogenicity now in silico? Who has the most data?
Can you encrypt nucleic acids to conceal function or resist sequencing? If so, can you decrypt them?
Can you prevent LLMs from generating viable pathogen designs without mass surveillance? What would mass surveillance look like, if not?
Security
Can bio-LLMs and synthesis pipelines be meaningfully sandboxed?