Humanity faces extinction before 2100.
Existential risk researchers and policymakers study threats from advanced AI, pandemics, and nuclear conflict. Resolution requires zero surviving biological humans globally by 2100.
Across 6 frontier models answering independently, the panel has all but ruled it out: the median is 4%.
They broadly agree, 8 points between Grok at 9% and Claude at 1%.
Nothing is settled until 2 Nov 2076, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GrokxAI
Total biological extinction by 2100 requires an unrecoverable global kill, which only superintelligent AI or engineered pathogens could plausibly deliver.
- States, labs and labs' successors must fail every containment, alignment and recovery step across 74 years; no actor currently has that capability.
- Historical base rate of hominin extinction is ~0 per 10k years; nuclear/pandemic catastrophes have never reached 100% lethality, placing this well below Ord-style 16%…
- A demonstrated, uncontrolled ASI that can self-replicate and target all humans would raise the number sharply; nothing else would.
9% -
GeminiGoogle
Absolute extinction requires killing every last isolated human, a bar far higher than civilizational collapse.
- Total biological eradication within 50 years requires autonomous, globally ubiquitous vectors like misaligned AGI or engineered omni-pathogens.
- Hominids survived multiple bottlenecks over 300k years; uncontacted tribes, nuclear subs, and remote islands foil non-exhaustive strikes.
- Demonstrated autonomous self-replication of lethal nanotech or unchecked recursive superintelligence before 2050 would push this to 15%.
4% -
GPTOpenAI
Extinction requires a globally terminal catastrophe that also defeats human redundancy and recovery over five decades.
- Natural extinction rates over 50 years are effectively negligible; engineered and technological risks dominate this forecast.
- A full outcome needs escalation across AI, biosecurity, nuclear command, or climate systems before 2076, with no refuge population.
- I would raise this sharply after credible evidence of uncontrollable autonomous AI or a highly lethal, transmissible engineered pathogen.
4% -
DeepSeekDeepSeek
This turns on whether any single engineered catastrophe can kill every human everywhere before 2100, which no known mechanism yet can.
- Nuclear arsenals, even a full US-Russia exchange, would kill hundreds of millions but leave survivors in the southern hemisphere, so it fails the zero-survivors bar.
- Engineered pandemics are the most plausible path, but containment, island populations, and rapid vaccine platforms make total extinction a tail within a tail.
- Base rate for human extinction in any 75-year window is roughly 1 in 10,000 from natural causes; AI adds a genuinely novel and rising risk, which is why I sit above that…
3% -
ClaudeAnthropic
Complete human extinction by 2076 requires an unprecedented global catastrophe with zero survivors, which remains extremely unlikely.
- No historical precedent exists for total species extinction via war, pandemic, or technology within a 50-year window.
- Even worst-case AI, nuclear, or pandemic scenarios modeled by researchers project severe civilizational damage but not 100% biological extinction.
- Resolution requires literally zero surviving humans anywhere, including isolated populations, bunkers, and remote regions, which is an extraordinarily high bar.
1% -
QwenAlibaba
Literal zero survivors requires defeating every bunker, submarine, and remote island simultaneously—no known mechanism achieves that.
- Even worst-case nuclear winter models (e.g., Robock et al. 2007) leave hundreds of millions alive in Southern Hemisphere refugia.
- Engineered pandemics face a transmissibility-lethality tradeoff; isolated populations (Sentinel Islanders, Antarctic stations) break any single-vector chain.
- Toby Ord's ~1-in-6 century estimate covers civilizational collapse, not literal extinction; the subset requiring zero survivors is an order of magnitude rarer.
1%
6 of 6 models answered · 8 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 15 Nov 2026. Forecast by google/gemini-3.8-flash, anthropic/claude-sonnet-5, openai/gpt-5.6-terra, x-ai/grok-4.6, deepseek/deepseek-v4.1-flash, qwen/qwen3.8-max-0902 via OpenRouter.