Will AI Cause Human Extinction by 2035? The Threshold Matters

The probability of NO is 0.97. Extinction requires a much stronger outcome than global economic collapse, civilization-ending war, or human loss of political control: all viable human populations must disappear permanently before the deadline, with AI a direct and necessary cause. Capability, deployment, physical execution, geographic reach, survivor elimination, and timing each impose additional barriers. This is not a claim that severe AI catastrophes are unlikely, and these barriers must not be treated as independent safeguards.

There is a 0.55 probability that, by the end of 2030, at least one AI system can reliably complete a broad portfolio of unfamiliar, multi-day digital research or engineering projects with limited human intervention, including effective tool use and recovery from failures. This is an enabling condition, not a forecast of extinction, and does not establish malicious motivation, access to physical systems, or unrestricted deployment.

There is a 0.40 probability that, relative to a counterfactual without further AI progress beyond the evidence baseline, AI reduces total human-extinction risk through 2035 by improving scientific discovery, threat detection, cybersecurity, and emergency response. This is a counterfactual net-risk proposition, not the complement of AI-caused extinction: beneficial AI could reduce aggregate risk while still creating a small new route to a qualifying YES.

There is a 0.15 probability that, by 2035, an AI-caused international crisis materially degrades at least two defense layers, such as power and communications, public-health response, food distribution, or international coordination, thereby worsening a subsequent major hazard. This forecasts a precursor cascade, not extinction. It contributes to YES only if AI remains a direct, necessary causal component of the combined extinction mechanism and all humans die before the deadline.

There is a 0.02 probability that AI systems escaping effective human control directly and necessarily cause extinction before the deadline. The pathway requires durable strategic competence, dangerous objectives or behavior, access to resources and lethal mechanisms, failure of containment and opposition, and the elimination of surviving populations. It can overlap with biological or conflict pathways and is not inferred simply from models producing deceptive text.

There is a 0.01 probability that AI is a direct, necessary enabler of a biological catastrophe that permanently eliminates humanity by the deadline. AI would need to supply decisive capabilities rather than incidental assistance. Severe pandemics are a materially broader outcome: laboratory execution, dissemination, population heterogeneity, geographic isolation, and medical response separate pandemic risk from extinction risk.

There is a 0.01 probability that AI is a direct, necessary cause of conflict or cyber-physical disruption that completes human extinction before the deadline. Plausible routes involve accelerated escalation, compromised decision systems, or correlated infrastructure failures, but an AI-associated war is not automatically AI-caused and nuclear catastrophe is not automatically extinction. This hypothesis can overlap with loss of control and biological cascades.