Biological Threat Exposure
Biological Threat Exposure
Overview
Section titled “Overview”Biological Threat Exposure measures society’s vulnerability to biological threats—including AI-enabled bioweapon development. Lower exposure is better—it means society has strong capacity to prevent, detect, and respond to both natural pandemics and engineered pathogens. Technological investment, governance frameworks, and the offense-defense balance all determine whether biosecurity capacity strengthens or weakens over time.
This parameter underpins:
- Prevention: Ability to stop dangerous biological research and acquisition
- Detection: Surveillance systems that identify outbreaks early
- Response: Capacity to develop countermeasures and contain threats
- Deterrence: Reducing incentives for biological attacks
Understanding biosecurity as a parameter (rather than just a “bioweapons risk”) enables symmetric analysis of both threats (AI-enabled offense) and defenses (AI-enabled detection), precise investment targeting across the biosecurity stack, trajectory assessment of whether the offense-defense balance is shifting, and threshold identification of minimum biosecurity capacity needed given advancing AI capabilities. This framing proves critical because advances in genetic engineering and synthetic biology, combined with rapid innovations in machine learning, are enabling novel biological capabilities that experts characterize as “offense-dominant and extremely difficult to defend against.”
Related analytical frameworks include the Bioweapons AI Uplift Model quantifying AI’s marginal contribution to biological threat capacity, the Bioweapons Attack Chain Model decomposing stages from ideation through deployment, and the Bioweapons Timeline Model projecting capability developments through 2030.
Parameter Network
Section titled “Parameter Network”Contributes to: Misuse Potential
Primary outcomes affected:
- Existential Catastrophe ↑↑↑ — AI-enabled bioweapons represent direct catastrophic threat
Current State Assessment
Section titled “Current State Assessment”Detection Capabilities
Section titled “Detection Capabilities”| System | Current Capability | Gap | Trend | 2025 Status |
|---|---|---|---|---|
| DNA synthesis screening | ~25% of dangerous sequences caught | 75% evade detection via AI design | Worsening (AI evasion) | Voluntary, patchy coverage |
| Metagenomic surveillance | Limited deployment (est. <5% of high-risk sites) | 95%+ of potential pathogens unmonitored | Slowly improving | CEPI’s 100 Days Mission driving investment |
| Clinical surveillance | Days-weeks to detect novel outbreaks | Speed insufficient for engineered pathogens (hours matter) | Stable | Legacy infrastructure constraints |
| Wastewater monitoring | Operational for SARS-CoV-2 in major cities | Limited to known pathogens; ~60% coverage gaps | Improving | Expanding to influenza, RSV |
Prevention Mechanisms
Section titled “Prevention Mechanisms”| Mechanism | Status | Effectiveness |
|---|---|---|
| DNA synthesis screening | Voluntary, patchy coverage | Microsoft research: AI evades 75%+ |
| Dual-use research oversight | National-level, inconsistent | Variable by jurisdiction |
| Biosafety lab standards | BSL system established | Compliance variable |
| Export controls | Focused on state programs | Less relevant to AI-enabled threats |
Response Capacity
Section titled “Response Capacity”The Coalition for Epidemic Preparedness Innovations (CEPI) has established the “100 Days Mission” aiming to compress vaccine development timelines from pathogen identification to regulatory approval to just 100 days—a dramatic reduction from the 12-18 month historical norm. This relies heavily on platform technologies, particularly mRNA vaccines which demonstrated during COVID-19 that they can proceed from genetic sequence to Phase 1 trials in 63 days (Moderna) and 42 days (Pfizer-BioNTech).
| Capability | Current Status | Target Capability | Improvement Trajectory |
|---|---|---|---|
| mRNA vaccine platforms | 42-63 days sequence-to-trial (proven COVID) | <30 days for novel pathogen response | Rapidly improving; trans-amplifying mRNA in development |
| Broad-spectrum antivirals | 2-3 effective families (e.g., molnupiravir) | Pan-coronavirus and pan-influenza coverage | Moderate R&D; limited investment |
| Medical countermeasures stockpiles | 40-60% below pre-COVID baselines (est.) | 120-day supply for 300M people | Slow rebuilding; budget constraints |
| Hospital surge capacity | 15-25% surge tolerance (regional variation) | 200-300% for pandemic response | Declining; staffing crisis |
What “High Biosecurity” Looks Like
Section titled “What “High Biosecurity” Looks Like”High biosecurity doesn’t eliminate all biological risk—it maintains robust capacity across prevention, detection, and response:
Key Characteristics
Section titled “Key Characteristics”- Robust screening: All DNA synthesis covered; AI-resistant detection
- Early warning: Metagenomic surveillance detects novel pathogens within days
- Rapid response: Vaccines and countermeasures in weeks, not years
- Coordination: International information sharing and joint response
- Deterrence: Attribution capability and consequence frameworks
Defense-in-Depth Stack
Section titled “Defense-in-Depth Stack”Factors That Decrease Biosecurity (Threats)
Section titled “Factors That Decrease Biosecurity (Threats)”AI-Enabled Offense
Section titled “AI-Enabled Offense”| Threat | Mechanism | Evidence | Uncertainty |
|---|---|---|---|
| Knowledge accessibility | AI synthesizes information for non-experts | GPT-4o3: 94th percentile virologist score | High - same accessibility aids defenders |
| Screening evasion | AI designs sequences that bypass detection | Microsoft: 75%+ evasion rate | Medium - detection AI also improving |
| Protocol assistance | AI troubleshoots lab procedures | Documented capability | Medium - also aids legitimate research |
| Combination effects | AI + cheap synthesis + automation | Converging trends | High - timeline uncertain |
Note: “Knowledge democratization” is dual-use—the same AI capabilities that could aid attackers also enable more researchers to work on countermeasures, vaccine development, and biosurveillance. See “AI for Defense” section below.
2025 Capability Assessments
Section titled “2025 Capability Assessments”The 2024 U.S. Intelligence Community Annual Threat Assessment explicitly warns: “Rapid advances in dual-use technology, including bioinformatics, synthetic biology, nanotechnology, and genomic editing, could enable development of novel biological threats.” This assessment reflects growing consensus that AI-enabled biology represents a distinct threat category requiring new defensive frameworks.
| Organization | Assessment | Action | Timeframe |
|---|---|---|---|
| OpenAI↗ | Next-gen models expected to hit “high-risk” classification for CBRN capabilities | Elevated biosecurity protocols; pre-deployment screening | 2025-2026 |
| Anthropic↗ | Activated ASL-3 for Claude Opus 4 over CBRN concerns | Additional safeguards; restricted access to biology tools | Activated Dec 2024 |
| National Academies | AI biosecurity monitoring and mitigation “urgently needed” | Comprehensive report with policy recommendations | March 2025 |
| Johns Hopkins/RAND | Convened expert workshop on hazardous capabilities of biological AI models | Developing risk assessment frameworks | June 2024 |
Structural Vulnerabilities
Section titled “Structural Vulnerabilities”| Vulnerability | Description | Mitigation Status |
|---|---|---|
| Voluntary screening | DNA synthesis screening not mandatory | Limited regulatory action |
| Screening gaps | Not all providers screen; benchtop synthesizers emerging | Growing concern |
| International coordination | No global biosecurity framework | Limited progress |
| Dual-use research | Legitimate research creates dangerous knowledge | Inconsistent oversight |
Escalating Capabilities
Section titled “Escalating Capabilities”| Capability | 2023 | 2025 | Trajectory |
|---|---|---|---|
| AI biology knowledge | High | Expert-level | Rapidly increasing |
| Synthesis planning assistance | Moderate | High | Increasing |
| Guardrail evasion | Variable | Low (frontier) / High (open-source) | Diverging |
| Integration with lab tools | Limited | Growing | Accelerating |
Factors That Increase Biosecurity (Supports)
Section titled “Factors That Increase Biosecurity (Supports)”Defensive Technologies
Section titled “Defensive Technologies”| Technology | Function | Status |
|---|---|---|
| Metagenomic surveillance | Detect any pathogen from environmental samples | Deployment expanding |
| mRNA vaccine platforms | Weeks from sequence to vaccine candidate | Proven with COVID |
| Far-UVC light | Safe disinfection of airborne pathogens | Emerging deployment |
| AI-enhanced detection | Pattern recognition for novel threats | Active development |
| Broad-spectrum antivirals | Work against multiple virus families | R&D ongoing |
Governance Mechanisms
Section titled “Governance Mechanisms”| Mechanism | Function | Status |
|---|---|---|
| Mandatory DNA screening | Universal coverage of synthesis providers | Proposed, not implemented |
| AI model biosecurity evaluations | Assess biological capability before release | Frontier labs implementing |
| International coordination | Share intelligence and response capacity | Limited |
| Dual-use research oversight | Review dangerous research proposals | Variable by country |
AI for Defense
Section titled “AI for Defense”AI democratization cuts both ways—the same capabilities that lower barriers for potential attackers also dramatically expand defensive capacity:
| Application | Benefit | Maturity | Democratization Effect |
|---|---|---|---|
| Pathogen detection | AI identifies novel sequences | Growing | Enables smaller labs and developing nations to participate in surveillance |
| Vaccine design | Accelerate candidate development | Proven | Open-source tools (AlphaFold, ESMFold) available to all researchers |
| Drug discovery | Find countermeasures faster | Active | AI reduces cost from $2.6B to potentially $100-500M per drug |
| Surveillance analysis | Process metagenomic data at scale | Developing | Cloud-based AI analysis accessible globally |
| Literature synthesis | Rapid review of pathogen research | Emerging | Non-specialists can quickly understand biosecurity literature |
| Threat anticipation | Model potential engineered pathogens | Research | Defenders can prepare countermeasures proactively |
The “democratization of biology” argument assumes attackers benefit more than defenders. However, defensive applications have larger user bases, more funding, regulatory support, and can operate openly—advantages that compound over time. The RAND finding that “wet lab skills remain the binding constraint” suggests knowledge democratization may benefit defenders more, since legitimate researchers already have lab access while potential attackers face persistent physical barriers.
Structural Improvements
Section titled “Structural Improvements”| Improvement | Status | Timeline |
|---|---|---|
| DNA synthesis database expansion | Growing | Ongoing |
| Secure DNA initiative | Proposed | Planning |
| International pathogen sharing | Post-COVID improvements | Slow progress |
| Pandemic preparedness treaties | WHO negotiations | Years away |
Why This Parameter Matters
Section titled “Why This Parameter Matters”Consequences of Low Biosecurity
Section titled “Consequences of Low Biosecurity”| Domain | Impact | Severity |
|---|---|---|
| Pandemic risk | Engineered pathogens could cause mass casualties | Catastrophic |
| Deterrence failure | Actors may attempt attacks if defenses are weak | High |
| Research chilling | Overreaction could harm beneficial biology | Moderate |
| Public health trust | Repeated failures erode cooperation | Moderate |
The Offense-Defense Balance
Section titled “The Offense-Defense Balance”Georgetown’s analysis characterizes advances in genetic engineering and synthetic biology as creating “destabilizing asymmetries” where offensive capabilities increasingly outpace defensive responses. However, this assessment remains contested. While biological design tools and generative AI can develop novel weapons that evade conventional detection, defensive AI systems face fundamental constraints—they must operate within legal frameworks while adversarial actors can break laws freely, creating structural advantage for attackers.
The balance hinges on three critical factors: (1) whether mandatory DNA synthesis screening can close the 75% evasion gap, (2) whether metagenomic surveillance deployment can achieve sufficient coverage (currently <5% of high-risk sites) to provide early warning, and (3) whether mRNA vaccine platforms can compress response times below the 100-day threshold. Expert probability estimates on long-term balance range from 25-45% favoring defense to 15-25% favoring offense, with 30-40% expecting ongoing contestation.
| Factor | Favors Offense | Favors Defense | Magnitude (1-5) | Notes |
|---|---|---|---|---|
| AI knowledge accessibility | ✓ | ✓ | 3 | Dual-use: aids both sides, but defenders have lab access advantage |
| Screening evasion capabilities | ✓ | 3 | 75% evasion rate with current systems; AI detection improving | |
| Synthesis cost reduction | ✓ | 2 | $1.09/base to <$1.01/base; affects defenders too (cheaper countermeasures) | |
| Legal/operational constraints | ✓ | 3 | Attackers unconstrained, but also unsupported and isolated | |
| mRNA vaccine platforms | ✓ | 4 | Proven 42-63 day timelines; improving further | |
| Metagenomic surveillance | ✓ | 4 | Game-changer if deployed at scale | |
| AI drug discovery | ✓ | 3 | Dramatically accelerates countermeasure development | |
| Defender resource advantage | ✓ | 4 | $100B+ legitimate biotech vs. isolated attackers | |
| Open collaboration | ✓ | 3 | Defenders share knowledge; attackers must work secretly | |
| Attribution capability | ✓ | 2 | Forensics improving; deters state actors | |
| Net balance | Contested | Contested | - | Expert estimates: 25-45% defense-favorable, 15-25% offense-favorable, 30-40% ongoing contestation |
Biosecurity and Existential Risk
Section titled “Biosecurity and Existential Risk”Toby Ord in The Precipice estimates 1 in 30 chance of existential catastrophe from engineered pandemics by 2100—second only to AI among anthropogenic risks. AI-enabled bioweapons could:
- Enable non-state actors to cause pandemic-level harm
- Reduce the barrier to attacks that previously required state resources
- Create novel pathogens beyond natural evolution
Trajectory and Scenarios
Section titled “Trajectory and Scenarios”Projected Trajectory
Section titled “Projected Trajectory”| Timeframe | Key Developments | Biosecurity Impact |
|---|---|---|
| 2025-2026 | Expert-level AI virology; ASL-3 activations | Stress testing defenses |
| 2027-2028 | Potential mandatory DNA screening; improved surveillance | Depends on governance |
| 2029-2030 | AI-designed countermeasures mature | Could shift balance to defense |
Scenario Analysis
Section titled “Scenario Analysis”| Scenario | Probability | Biosecurity Outcome | Key Indicators (by 2028) | Offense-Defense Balance |
|---|---|---|---|---|
| Defense Strengthens | 35-45% | Surveillance and vaccines outpace offense; AI democratization benefits defenders more | Mandatory DNA screening implemented; metagenomic coverage >40%; <50-day vaccine response proven; open-source bio-defense tools proliferate | Defense +2 |
| Contested Balance | 35-45% | Ongoing cat-and-mouse; both sides improve; no major incidents | Voluntary screening expands; selective surveillance deployment; 60-90 day vaccine timelines; incremental progress on both sides | Neutral |
| Offense Advantage | 10-20% | AI-enabled attacks exceed defense capacity | Screening remains voluntary; surveillance <10% coverage; 100+ day responses; successful engineered pathogen release | Offense +2 |
| Catastrophic Incident | 5-10% | Major biological event forces reactive global response | Engineered outbreak with >100K casualties; emergency treaty negotiations | Depends on response |
Note: The “Defense Strengthens” and “Contested Balance” scenarios together account for 70-90% of probability mass. Catastrophic outcomes remain possible but are not the most likely trajectory given current defensive investments and the structural advantages defenders hold (resources, collaboration, legitimacy).
Critical Dependencies
Section titled “Critical Dependencies”| Factor | Importance | Current Status |
|---|---|---|
| DNA synthesis screening coverage | Very High | Incomplete |
| AI model biosecurity evaluation | High | Frontier labs only |
| Metagenomic surveillance deployment | High | Limited |
| International coordination | Very High | Weak |
Key Debates
Section titled “Key Debates”Are AI-Bioweapons Overhyped?
Section titled “Are AI-Bioweapons Overhyped?”This debate centers on conflicting 2024-2025 evidence. RAND Corporation’s January 2024 study concluded that “current artificial intelligence does not meaningfully increase risk of a biological weapons attack” by non-state actors, finding wet lab skills remain the binding constraint. This finding has held up through 2025 despite advancing AI capabilities.
Higher concern view (25-40% expert probability):
- AI lowers knowledge barriers (GPT-4o3: 94th percentile virologist performance)
- Screening systems currently catch only 25% of dangerous sequences
- Historical non-occurrence may reflect luck or lack of motivated actors
- Future AI capabilities may overcome current constraints
Lower concern view (30-45% expert probability):
- RAND study found no significant AI uplift for current or near-term models
- Wet lab skills remain the binding constraint (equipment, technique, scale-up)—AI doesn’t help here
- Existing scientific literature already contains dangerous information; AI adds little marginal risk
- Natural pandemics pose greater near-term risk; resources better spent on general preparedness
- AI democratization benefits defense more (see above)—larger user base, more funding, open collaboration
- No successful AI-enabled bioattacks despite years of AI availability
Balanced/pragmatic view (30-40% expert probability):
- Risk is genuine but probability bounds remain wide (5-25% for catastrophic event by 2050)
- Prudent to invest in defense while avoiding overreaction that chills beneficial biology research
- Defensive investments (surveillance, vaccines) provide value against both natural and engineered threats
- The “democratization helps attackers” framing ignores that defenders also benefit from AI accessibility
- 2025-2027 capability trajectory will provide crucial evidence; current alarmism may be premature
Mandatory vs. Voluntary Screening
Section titled “Mandatory vs. Voluntary Screening”Mandatory screening:
- Closes gaps in coverage
- Levels competitive playing field
- Enables enforcement
Voluntary approach:
- Less bureaucratic burden
- Industry innovation
- Avoids regulatory capture
Related Pages
Section titled “Related Pages”Related Risks
Section titled “Related Risks”- Bioweapons — Comprehensive analysis of AI-enabled biological threats and attack vectors
- Dual-Use Research Risks — Legitimate research creating dangerous capabilities
Related Models
Section titled “Related Models”- Bioweapons AI Uplift Model — Quantifies AI’s marginal contribution to bioweapons capability (finding 1.3-2.5x uplift for non-experts)
- Bioweapons Attack Chain Model — Decomposes stages from ideation through deployment with probability estimates
- Bioweapons Timeline Model — Projects when AI capabilities cross critical thresholds through 2030
Related Interventions
Section titled “Related Interventions”- Responsible Scaling Policies — Industry frameworks for managing AI biosecurity risks
Related Parameters
Section titled “Related Parameters”- Cyber Threat Exposure — Parallel analysis of digital security offense-defense balance
- Societal Resilience — Society’s broader capacity to recover from catastrophic shocks
- International Coordination — Global governance affecting biosecurity treaties and information sharing
Sources & Key Research
Section titled “Sources & Key Research”Government and Policy Assessments
Section titled “Government and Policy Assessments”- U.S. Intelligence Community (2024): “Annual Threat Assessment” — Identifies rapid advances in bioinformatics, synthetic biology, and genomic editing as enabling novel biological threats
- National Academies (2025): “The Age of AI in the Life Sciences” — Comprehensive report recommending urgent monitoring and mitigation frameworks
- NATO (2024): Adopted strategy to guide biotechnology development for defensive purposes
- OpenAI↗ biosecurity evaluations and capability assessments
- Anthropic↗ ASL-3 documentation and CBRN threshold activation
Academic Research (2024-2025)
Section titled “Academic Research (2024-2025)”- RAND Corporation (January 2024): “Current Artificial Intelligence Does Not Meaningfully Increase Risk of a Biological Weapons Attack” — Empirical study finding wet lab skills remain binding constraint for current LLMs
- Georgetown Journal of International Affairs (2025): “Deterrence in the Age of Weaponizable Biotechnology” — Analysis characterizing genetic weapons as “offense-dominant and extremely difficult to defend against”
- Future of Life Institute (2024): “AI and Chemical & Biological Weapons” — Comprehensive threat assessment and policy recommendations
- Johns Hopkins Center for Health Security & RAND (June 2024): Joint workshop on hazardous capabilities of biological AI models trained on biological data
- Microsoft Research: AI screening evasion techniques achieving 75%+ bypass rates
Vaccine and Response Technology (2024)
Section titled “Vaccine and Response Technology (2024)”- Coalition for Epidemic Preparedness Innovations (CEPI) (2024): “Fast-Tracking Vaccine Manufacturing: Rapid Response Framework for the 100 Days Mission” — Establishes framework to compress vaccine development to 100 days
- CEPI (March 2024): “New Research on Trans-Amplifying mRNA Vaccines” — $1M funding for next-generation self-amplifying vaccines requiring lower doses
- Nature Signal Transduction (2024): “Progress and Prospects of mRNA-based Drugs in Pre-clinical and Clinical Applications” — Comprehensive review of mRNA platform advances
- Virology Journal (2025): “Revolutionizing Immunization: A Comprehensive Review of mRNA Vaccine Technology” — Analysis of rapid response capabilities and manufacturing innovations
International Governance
Section titled “International Governance”- Biological Weapons Convention (2025): 50th anniversary approaching; treaty review considering AI and synthetic biology governance challenges
- Bulletin of the Atomic Scientists (November 2024): “What Will Be the Impact of AI on the Bioweapons Treaty?” — Analysis of treaty adaptation requirements
- World Health Organization (2024): R&D Blueprint with updated priority and prototype pathogens for pandemic preparedness
Industry Initiatives
Section titled “Industry Initiatives”- Frontier Model Forum↗ biosecurity working group developing shared evaluation standards
- International Gene Synthesis Consortium (IGSC): Voluntary screening protocols and coordination
- Nuclear Threat Initiative (NTI): Biosecurity program focusing on DNA synthesis governance
What links here
- Human-Caused Catastrophescenariokey-factor
- Misuse Potentialrisk-factorcomposed-of
- Bioweapons Attack Chain Modelmodelmodels
- AI Uplift Assessment Modelmodelaffects