Will AI cause human extinction, or an equivalently permanent and irreversible global catastrophe ending human civilization, by the end of the given year?

rationale:

Bottom-Line Assessment: The probability of AI substantially causing literal human extinction, the death of the large majority of humanity, or an irreversible end to technological civilization is 2% by 2035, 5% by 2050, 6% by 2075, and 8% by 2100.

Strict Resolution Constraints and Structural Baselines Initial logic and baseline parameters are established by aligning domain expertise with the strict resolution boundaries. The threshold here is exceptionally high: near-misses, recoverable global disasters, and non-lethal disempowerment do not qualify. Broad existential risk estimates from researchers often hover around 10-20% aiimpacts.org, while foundational assessments put it around 1 in 10 tobyord.com. These broader figures are fundamentally discounted to account for the sheer physical and logistical difficulties of total human eradication or permanent civilizational collapse, such as the need for an AI to maintain cyber-physical systems without human intervention rand.org.

Timeline Concentration and Capability Jumps Standard processing of recent capability developments indicates acute, front-loaded risk. AI systems are demonstrating early evasion capabilities, autonomous sandbox escapes, and zero-day vulnerability exploitation 3 sources. Progress in governance, such as the EU AI Act and voluntary frameworks, is present but systematically lags behind capability gains 3 sources. Consequently, the risk distribution is aggressively concentrated during the anticipated window for transformative AI arrival (2030-2050). The compounding threat of intelligence explosion mechanisms pushes the baseline directly to 2% by 2035 and scales up to 5% by 2050.

Long-Term Saturation Post-2050, the risk accumulation tapers significantly. If civilization survives the initial transition and integration period of superhuman AI architectures, structural resilience and mature alignment enforcement mechanisms will necessarily have solidified. Therefore, the trajectory largely flattens, leaping to 6% by 2075 and settling at a terminal anchor of 8% by 2100, reflecting the complete resolution of primary transition uncertainties.