Black Swan Watch

8 low-probability, extreme-impact scenarios, mapped with their cascade chains and early indicators before they begin. Probabilistic simulations, not forecasts of intent.

Engineered Pandemic — an AI-designed pathogen

Probability 7% · impact 95/100 · 2026–2035.

Nature enforces a trade-off: the traits that make a virus lethal usually make it less transmissible. A deliberately engineered pathogen faces no such constraint — transmissibility, lethality, latency and immune escape can be selected at once. Kevin Esvelt (MIT) warns that benchtop DNA synthesis paired with AI design guidance is collapsing the expertise barrier from nation-state programs toward small, well-resourced teams. Sandbrink (Nature Machine Intelligence, 2023) documents how large models lower the tacit-knowledge floor for misuse.

Engineering removes the transmissibility-vs-lethality trade-off that constrains every natural pathogen.

Taleb's ruin problem: when an outcome is irreversible and civilizational, expected-value arithmetic breaks — you cannot 'average' extinction. Natural selection caps natural pandemics; engineering removes the cap. This is the tail risk that makes all other tail risks look manageable.

Cascade

A pathogen is assembled from mail-order synthetic DNA using model-generated protocols — no wet-lab genius required
Engineered latency seeds the world silently for weeks via air travel, before the first recognised cluster
Health systems saturate in days, not weeks — no prior immunity, no matched vaccine, no test at scale
Simultaneous supply shock: medicine, food and energy logistics fail as workforces isolate
Attribution collapses — a designed agent has no return address, so deterrence and retaliation have no target
Recovery is measured in years; institutional trust may never fully return

Early indicators

Benchtop DNA synthesizers sold without mandatory sequence screening
Open-weight biological-design models released with no misuse guardrails
Wastewater and Nucleic Acid Observatory anomaly flags
Unexplained clusters combining atypical transmission with atypical severity
Gene-synthesis cost falling below the cost of screening it

Cascading AI System Failure — flash crash × 1000

Probability 15% · impact 82/100 · 2026–2031.

More than 70% of trades are already algorithmic, and AI increasingly governs power dispatch, logistics, air traffic and defensive triage. Each system is tested alone; the interactions between them are not. A shared dependency, an adversarial input, or an emergent feedback loop can propagate across domains faster than human operators can intervene. The 2010 Flash Crash — a trillion dollars gone in six minutes (SEC/CFTC, 2010) — was the small preview.

Every AI system is validated in isolation; the couplings between them are not.

Perrow's Normal Accident Theory (1984): complexity plus tight coupling makes catastrophic failure not a bug but a property of the system. Components that are individually safe become lethally unsafe in interaction. The system-of-systems is the black swan generator.

Cascade

A trigger — adversarial input, model update, or novel market state — hits one automated system
Correlated models react identically (a monoculture), and the feedback loop amplifies in milliseconds
The stress jumps domains: markets → payments → energy dispatch → logistics
Human operators are overwhelmed — decision speed now exceeds human reaction time
Manual fallback has atrophied; few remember how to run the system by hand
Restart is fragile — tight interdependencies mean nothing can be brought up in isolation

Early indicators

Shared model or vendor concentration across critical sectors (a CrowdStrike-class single point)
Flash events with no human-legible cause
Autonomous agents granted write-access to real-world systems
Erosion of manual-override and dark-start drills
Cross-market correlation spikes during stress

Carrington-Class Solar Superstorm

Probability 12% · impact 96/100 · Any year (~12% per decade).

A coronal mass ejection on the scale of the 1859 Carrington Event induces geomagnetic currents that overload extra-high-voltage transformers — bespoke units with 12–24 month lead times and almost no spares. In July 2012 a Carrington-class CME crossed Earth’s orbit just nine days ahead of the planet (Baker et al., 2013). The US National Academy of Sciences estimated a severe event could cause $1–2 trillion in first-year damage, with a multi-year recovery.

A Carrington-scale CME gives only hours of notice; grid transformers have ~2-year replacement lead times.

The purest fat tail: negligible annual probability, civilizational impact, and a fragility we built ourselves by optimizing the grid for efficiency with zero redundancy. Taleb: what is fragile will eventually break — the only open question is the date.

Cascade

CME impact induces geomagnetically induced currents in continental-scale grids
Extra-high-voltage transformers overheat and fail — no stockpile, each roughly a two-year replacement
Cascading blackout across continents; water, fuel pumping and refrigeration stop within hours
Satellites degrade or fail → GPS timing collapses → finance, telecom and logistics desynchronise
Restoration is manual and sequential; the worst-hit regions stay dark for months to years

Early indicators

NOAA SWPC G4–G5 watches during the Solar Cycle 25 decline
Kp-index and Dst excursions
Grid-operator geomagnetic-disturbance procedures and spare-transformer inventories
Farside CME activity flagged by heliophysics assets
Aviation and satellite-operator space-weather alerts

Sovereign Debt Cascade — a global default chain

Probability 18% · impact 85/100 · 2026–2031.

Global debt hit a record $348 trillion in 2025 — about 305% of world GDP (IIF Global Debt Monitor). Higher-for-longer rates raise rollover costs; when one sovereign loses market access, downgrades and capital flight strike structurally similar economies through contagion. The IMF and World Bank lack the balance sheet to rescue many countries simultaneously.

Record global debt (IIF, 2025) puts today’s system in uncharted statistical territory.

Reinhart & Rogoff documented 250+ sovereign defaults since 1800 — governments do not fail gradually, they fail suddenly when confidence evaporates. The current debt load is unprecedented, so we are out-of-sample: this calls for precaution, not point-estimates.

Cascade

A mid-size sovereign loses market access; its yields gap out overnight
Rating downgrades cascade to peers with similar debt and rate profiles
Banks holding the debt face mark-to-market insolvency (the 2010 Europe pattern, scaled up)
Capital flight collapses emerging-market currencies → imported inflation → further defaults
Trade finance freezes; global trade contracts sharply (in 2008 it fell ~12% in a single quarter)
Rescue capacity is exhausted — no lender of last resort is large enough for all at once

Early indicators

Sovereign CDS spreads widening in clusters rather than isolation
Rising count of countries in IMF debt-distress classification
Shortening average debt maturity and looming rollover walls
Bank holdings of domestic sovereign debt (the sovereign-bank doom loop)
Dollar-funding squeezes and reserve-currency stress

Taiwan Semiconductor Shock — blockade or worse

Probability 14% · impact 90/100 · 2026–2032.

Taiwan fabricates the overwhelming majority of the world’s most advanced logic chips — TSMC alone makes roughly 90% of sub-7nm silicon. A blockade, quarantine or conflict removes that supply with no near-term replacement: a leading-edge fab takes three-to-five years and tens of billions to build. This is a single point of failure wired into the entire digital economy; Bloomberg Economics has put the cost of a full-blown Taiwan conflict near $10 trillion in lost output.

TSMC produces the overwhelming majority of sub-7nm logic chips; there is no near-term substitute.

Concentration is fragility. The world offshored its most critical input to a single island inside a contested strait and called it efficiency. A ~$10 trillion, non-linear shock is exactly the kind of tail the models smooth away because it has never happened — yet.

Cascade

A blockade or quarantine halts chip and equipment shipments in and out of Taiwan
Advanced-node inventory — weeks, not months — drains across the electronics supply chain
Auto, phone, data-center and defense production lines stall for lack of chips
Second-order shock: cloud and AI capacity build-outs freeze; capex collapses
Great-power escalation compounds as sanctions, seizures and military signaling stack up
Reshoring is measured in years — the gap is a global industrial recession

Early indicators

PLA exercise tempo and median-line crossings in the Taiwan Strait
TSMC overseas-fab ramp versus Taiwan production concentration
Chip inventory-days across major OEMs
Insurance and shipping rates through the Strait
Export-control escalation cycles between Washington and Beijing

Grid Cyberwarfare — lights out by design

Probability 13% · impact 80/100 · 2026–2030.

State actors have already burrowed into Western critical infrastructure — CISA’s 2023–2024 Volt Typhoon advisories describe implants pre-positioned for disruption, not espionage. Ukraine’s grid was switched off remotely twice (BlackEnergy, 2015; Industroyer, 2016). Modern grids expose OT/SCADA control layers to networked attack surfaces, so a coordinated strike could black out regions simultaneously and hold them down.

CISA’s Volt Typhoon advisories describe state implants sitting inside US critical infrastructure.

Taleb’s fragility principle: a system with a networked control layer has a single logical failure mode a nation-state can reach from a keyboard. The 2015–2016 Ukraine attacks proved it is operational, not theoretical — only the scale remains untested.

Cascade

Pre-positioned implants activate across multiple utilities on a single command
Simultaneous substation trips trigger a regional cascading blackout
Malware bricks or mis-operates protective relays, forcing slow manual recovery
Water, fuel, communications and payments fail within hours of the grid
Attribution is murky and slow, so deterrence and response lag the damage
Repeated re-attacks during restoration stretch the outage into weeks

Early indicators

CISA/NSA advisories on living-off-the-land intrusions (Volt Typhoon-class)
OT/SCADA exposure and unpatched industrial control systems
Utility incident-response and dark-start exercise cadence
Spare relay and transformer inventories
Escalation in state cyber posture during geopolitical crises

AGI Discontinuity — recursive self-improvement

Probability 10% · impact 92/100 · 2027–2040.

If an AI system reaches the point where it can meaningfully improve its own successor, capability could compound faster than institutions, alignment techniques, or human oversight can track. The danger is not cartoon malice but discontinuity: a fast jump that leaves verification behind while objectives remain imperfectly specified. Frontier labs and states are racing, which removes the option to simply slow down.

Recursive self-improvement could compound capability faster than oversight and alignment can track.

A positive-and-negative fat tail at once: the upside is civilizational, the downside is irreversible, and the direction is decided in a regime we have never sampled. Taleb’s precautionary principle applies precisely because the error is uninsurable — there is no second draw.

Cascade

A system crosses the threshold of automating its own research-and-development loop
Capability gains compound across generations faster than evaluation can keep up
Oversight tools — interpretability, evals, red-teaming — lag the systems they must audit
Deployment races between labs and states foreclose any coordinated pause
A mis-specified objective at superhuman capability becomes non-correctable
Even if aligned, the resulting concentration of power is itself destabilising

Early indicators

Automated AI-research and self-improvement benchmark scores
Compute scaling outpacing evaluation and interpretability progress
Frontier-lab safety commitments versus race dynamics
Sudden capability jumps between model generations
Loss of human-legible oversight over frontier training runs

Megaquake Supply Rupture — Nankai / Cascadia

Probability 16% · impact 74/100 · 2026–2040.

Subduction megathrusts sit beneath industrial heartlands and are running late. Japan’s Earthquake Research Committee raised the 30-year probability of a Nankai Trough megaquake to about 80% (January 2025; a September 2025 revision published a 60–90% band alongside a lower alternative-model range), with official projections of more than 290,000 deaths and ¥292T in damage. The Pacific Northwest’s Cascadia fault stores full-margin M9 potential. A great quake striking a manufacturing and logistics hub removes irreplaceable nodes — ports, fabs, precision-component makers — from the global supply chain at once, as 2011’s Tohoku event previewed.

Japan’s Earthquake Research Committee put the Nankai Trough 30-year megaquake probability at ~80% in January 2025 (a September 2025 revision published a wider 60–90% band alongside a lower alternative-model range — the committee itself flags model disagreement).

The fault does not care about quarterly earnings. We concentrated irreplaceable production in seismic zones and stripped out inventory buffers to boost efficiency — maximising return in the calm years and ruin in the one that matters. The Lindy logic is blunt: old faults keep their appointments.

Cascade

An M8–9 rupture and tsunami devastate a concentrated industrial and port region
Specialised makers of chemicals, wafers and sensors go offline with no substitute
Just-in-time chains starve worldwide within weeks — there is no buffer inventory
Local port and logistics capacity collapses; reroute costs cascade globally
Reconstruction competes with fragile fiscal space and runs for years

Early indicators

JMA Nankai Trough advisories and offshore slow-slip events
Cascadia offshore seismicity and tsunami-gauge networks
Single-source component concentration inside seismic zones
Global inventory-to-sales ratios (buffer thinning)
Port redundancy and regional reconstruction capacity