Pandemic Risk & H5N1
After 2020 the incentive runs toward alarm, which makes a low sealed number the harder and more useful thing to publish. The threshold below is specific: sustained human-to-human transmission, not case counts.
Live projections
The H5N1 clade 2.3.4.4b panzootic is the widest avian-influenza event ever recorded: global spread in wild birds, spillovers into an unprecedented range of mammals, and — since 2024 — a sustained outbreak in US dairy cattle with dozens of confirmed, mostly mild human cases among farm workers and one US death reported in January 2025. What has NOT happened is the only thing that matters for this claim: no confirmed chain of sustained human-to-human transmission, anywhere, ever, for H5N1. The virus needs a small set of receptor-binding and polymerase changes to transmit efficiently between humans; surveillance watches exactly those mutations. Historical base rates keep the sealed number low — H5N1 has sat at the threshold since 1997 without crossing — but the exposure surface (cattle, farm workers, raw-milk chains) has never been this broad. Low probability, ruin-class consequence: the classic tail.
Wrong if: Resolves CORRECT if, by 2028-12-31, WHO officially confirms at least one cluster of sustained human-to-human transmission of influenza A(H5) (three or more transmission generations), or declares a Public Health Emergency of International Concern for A(H5). Resolves INCORRECT if neither occurs by that date. Sporadic spillover cases with no sustained chains do not count.
Window: 2026-2028.
Acute food insecurity has risen six years straight — 295M people across 53 countries in 2024 (GRFC 2025) — before any simultaneous harvest failure. The system's fragility is structural: Russia and Belarus control ~40% of potash exports, a handful of basins irrigate the surplus, and export bans propagate faster than harvests recover.
Wrong if: the Global Report on Food Crises records declining acute food insecurity for two consecutive years before 2030, with no simultaneous two-breadbasket production failure.
Window: 2026-2030.
Tail scenarios
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.
Other subjects on the record
Every claim above is also listed in the full projection registry and the calibration ledger. Scoring is explained in how a prediction is made.