S-tier only

Problems humanitycannot afford to lose.

S-tier is not the top percentile of a long list. It is a consequence class: a credible path to human extinction or permanent loss of civilization's future. The ranking below separates the existential mechanism from the concrete defenses a founder, researcher, funder, or institution can build now.

4 threat classes12 concrete quests0 invented probabilities

The admission test

1. Species-wide consequenceA credible mechanism reaches every population or permanently destroys humanity's future.

2. Evidence, with uncertaintyThe mechanism is sourced. Unknown probability is shown as unknown, never converted into fake precision.

3. Actionable defenseThere is a measurable bottleneck where additional engineering, science, capital, or coordination can change the outcome.

S1

AI control

Loss of control over advanced AI

Failure mechanism

Highly capable systems could evade oversight, acquire resources, resist shutdown, or give a small group irreversible power over everyone else.

What evidence says

Current systems cannot cause loss of control. Relevant capabilities are improving, while the probability and timing of future loss of control remain deeply uncertain.

Victory condition

No system capable of defeating human oversight is deployed without demonstrated containment, independent evaluation, and a reliable path to shutdown.

Rate-limiting bottleneck

We can measure benchmark performance faster than we can measure deceptive behavior, long-horizon autonomy, situational awareness, and whether safeguards survive adversarial pressure.

Buildable quests

proof, not aspiration
01

The control plane for AI agents

A runtime permission, isolation, tripwire, and immutable-audit layer for agents operating computers, codebases, laboratories, or financial systems.

PROOF: An evaluated agent cannot exceed a declared capability envelope without a logged, human-authorized escalation.

02

Independent frontier evaluations

A technically credible third-party lab that tests autonomy, oversight evasion, cyber, and biological capabilities before deployment.

PROOF: Major deployers and regulators use reproducible external evaluations as a real release gate.

03

Verifiable compute and model provenance

Hardware-backed records showing which model ran, with what weights, permissions, tools, and compute budget.

PROOF: Auditors can verify high-risk training and deployment claims without receiving model weights or private user data.

What you can do

Open data

Build

Startups, labs, open work

  • Nonprofit

    Redwood Research

    Pioneered AI control research on keeping possibly misaligned AI systems safe.

    Checked 2026-11-01

  • Nonprofit

    METR

    Evaluates frontier AI models for dangerous autonomous capabilities.

    Checked 2026-11-01

  • Nonprofit

    Apollo Research

    Researches and evaluates deceptive behavior and scheming in AI systems.

    Checked 2026-11-01

  • Government · UK

    UK AI Security Institute

    UK government institute testing frontier AI and funding alignment and control research.

    Checked 2026-11-01

  • Nonprofit

    FAR.AI

    Runs AI safety research and field-building events.

    Checked 2026-11-01

  • Startup

    Goodfire

    Builds interpretability tools for understanding and editing neural networks.

    Checked 2026-11-01

  • Nonprofit

    Transluce

    Builds open tools for understanding and overseeing AI systems.

    Checked 2026-11-01

  • Nonprofit

    MATS

    Research mentorship program that trains new AI alignment researchers.

    Checked 2026-11-01

Fund

Grants, prizes, funders

Buy

Products to use or back

Learn

Explainers and papers

Primary evidence: International AI Safety Report 2026

Discoveries

All 2
S2

Pandemic defense

Engineered or extreme pandemics

Failure mechanism

A highly transmissible pathogen can replicate, cross borders before symptoms appear, and attack people, food systems, or critical institutions everywhere at once.

What evidence says

The last natural pandemic demonstrated global propagation. Genetic engineering, synthesis, and AI-enabled biological tools add misuse and accident pathways that require stronger defenses.

Victory condition

Detect any exponentially growing pathogen within days, contain it locally, and manufacture effective countermeasures at global scale before the first major wave.

Rate-limiting bottleneck

Surveillance is fragmented, synthesis screening is not universal, and vaccine design is much faster than regulated manufacturing and distribution.

Buildable quests

proof, not aspiration
01

Pathogen-agnostic early warning

Continuous metagenomic monitoring of wastewater, transport hubs, farms, and building air that alerts on unknown biological growth.

PROOF: A novel pathogen is detected and localized before hospitals observe the first large clinical wave.

02

Universal nucleic-acid screening

Privacy-preserving screening infrastructure for synthesis providers, benchtop devices, and high-risk biological design workflows.

PROOF: Dangerous orders and design patterns are intercepted across participating providers without exposing legitimate customer sequences.

03

Days-not-months biomanufacturing

Distributed, reconfigurable manufacturing with prevalidated processes for vaccines, diagnostics, and broad-spectrum countermeasures.

PROOF: A validated countermeasure moves from pathogen sequence to globally meaningful production in weeks.

What you can do

Open data

Build

Startups, labs, open work

  • Nonprofit

    SecureBio

    Builds pandemic early-warning and AI-biosecurity defenses, including DNA synthesis screening research.

    Checked 2026-11-01

  • Nonprofit

    Nucleic Acid Observatory (now SecureBio Early Detection)

    Runs metagenomic sequencing of wastewater and air to detect novel or engineered pathogens early.

    Checked 2026-11-01

  • Nonprofit

    Blueprint Biosecurity

    Develops roadmaps and projects for far-UVC, next-generation PPE, and other pandemic-proofing technologies.

    Checked 2026-11-01

  • Nonprofit

    IBBIS (International Biosecurity and Biosafety Initiative for Science)

    Develops the open Common Mechanism tool and standards for screening DNA synthesis orders.

    Checked 2026-11-01

  • Nonprofit

    SecureDNA

    Offers free, privacy-preserving cryptographic screening of DNA synthesis orders for hazardous sequences.

    Checked 2026-11-01

  • Startup · US

    Biobot Analytics

    Turns wastewater sampling into community-level pathogen surveillance data.

    Checked 2026-11-01

  • Startup

    Far UV Technologies

    Makes far-UVC lamps intended to inactivate airborne pathogens in occupied rooms.

    Checked 2026-11-01

  • Nonprofit

    Johns Hopkins Center for Health Security

    Researches and advises governments on preventing and responding to pandemics and biological threats.

    Checked 2026-11-01

Fund

Grants, prizes, funders

Buy

Products to use or back

  • Product

    Ushio Care222 far-UVC

    Commercial far-UVC lamp modules for disinfecting air in occupied indoor spaces.

    Checked 2026-11-01

  • Product

    Far UV Technologies Krypton lamps

    Ready-to-install far-UVC fixtures for offices, clinics, and schools.

    Checked 2026-11-01

  • Product · US only

    Vaccines.gov

    Finds nearby flu, COVID-19 and other vaccines (US-only).

    Checked 2026-11-01

Learn

Explainers and papers

Primary evidence: WHO laboratory biosecurity guidance

Discoveries

All 2
S3

Nuclear stability

Nuclear escalation and nuclear winter

Failure mechanism

Warning errors, cyberattack, miscalculation, or war could escalate into mass detonations. Firestorms could then disrupt climate and collapse global food production.

What evidence says

The extinction probability is uncertain, but modeled large-war scenarios produce civilization-scale famine. Existing arsenals make this an active hazard, not a future invention.

Victory condition

Prevent accidental or intentional escalation while making food, communications, energy, and knowledge resilient to abrupt global catastrophe.

Rate-limiting bottleneck

Nuclear decision systems are opaque and politically constrained, while most resilience planning assumes sunlight, trade, power, and normal agriculture continue.

Buildable quests

proof, not aspiration
01

Verification without strategic exposure

Sensors and cryptographic protocols that verify limits, dismantlement, and treaty compliance without revealing sensitive weapons data.

PROOF: Adversarial states can verify a safety claim that neither side could previously trust.

02

Crisis communication that survives attack

Authenticated, redundant communication and decision-support infrastructure designed to reduce false alarms and preserve de-escalation options.

PROOF: Leaders retain a trusted communication path and independent warning checks during a severe cyber or infrastructure failure.

03

Food without normal sunlight

Scalable foods based on fermentation, stored feedstocks, resilient crops, and industrial conversion processes that work after abrupt sunlight reduction.

PROOF: A region can supply minimum nutrition for years without depending on normal harvests or global food trade.

What you can do

Open data

No checked action is listed for this yet.

Build

Startups, labs, open work

None yet.

Fund

Grants, prizes, funders

None yet.

Buy

Products to use or back

None yet.

Learn

Explainers and papers

None yet.

Primary evidence: Nature Food nuclear-winter model

Discoveries

All 2
S4

Civilization continuity

Single-planet catastrophic exposure

Failure mechanism

Asteroids, long-period comets, supervolcanoes, or an unknown global shock can destroy modern civilization while every human population and supply chain shares one planet.

What evidence says

NASA has demonstrated asteroid deflection, but detection and response are incomplete. A self-sustaining second biosphere does not yet exist.

Victory condition

Detect and divert impactors, preserve an independent recovery capability on Earth, and ultimately maintain a self-sustaining human biosphere beyond Earth.

Rate-limiting bottleneck

Long-period objects can arrive with limited warning, continuity systems are rarely tested, and off-world settlements remain dependent on terrestrial resupply.

Buildable quests

proof, not aspiration
01

Complete the impactor map

Detection, orbit determination, and rapid characterization for hazardous asteroids and comets, including objects difficult to observe from Earth.

PROOF: Every object above a defined damage threshold is found with enough warning time to mount a response.

02

Planetary defense as a repeatable capability

Rapid-response launch, reconnaissance, and deflection systems that turn a successful demonstration into standing infrastructure.

PROOF: Humanity can launch a characterized deflection mission inside the minimum warning window.

03

A tested civilization recovery stack

Distributed energy, tools, seed and microbial libraries, manufacturing instructions, communications, and institutions designed for independent recovery.

PROOF: A geographically isolated site can maintain essential industry and restore lost capabilities without functioning global supply chains.

What you can do

Open data

No checked action is listed for this yet.

Build

Startups, labs, open work

None yet.

Fund

Grants, prizes, funders

None yet.

Buy

Products to use or back

None yet.

Learn

Explainers and papers

None yet.

Primary evidence: NASA Planetary Defense

The focus rule

Everything else is supporting infrastructure.

Energy, climate resilience, institutions, scientific productivity, and demographic health matter here when they measurably reduce one of these four risks. They are not relabeled existential merely because they are important.

Find your S-tier fit