is your estimated lifetime chance of dying from AI-related causes
These odds aren't fixed. See what you can do ↓Whose assumptions?
You're now using your own custom settings. You can pick a worldview or reset below.
Nobody knows the true odds of AI catastrophe. The answer depends on who you trust. Pick a worldview to re-run the model on that person or group's published predictions, or set any assumption yourself below the chart.
A caveat on these worldviews
Each is our attempt to map a person or group's public statements onto this model's structure, which they had no hand in designing. The figures shown might differ from what they would actually endorse. Where related opinions have been stated publicly, we cite them so you can compare. Our model depends on the estimated danger of AI and when high-level AI capabilities are reached, so we try to include opinions about both. Where timeline estimates are unavailable, we use defaults based on the current ~4 month rate doubling of METR task-completion time horizons, assuming minimal government action to slow things down. (You could change this ↓)
Lifetime probability of death, by cause
The everyday causes (heart disease, cancer, accidents…) come from US mortality records (SSA & CDC) and differ by country. The AI-risk rows are global estimates.
Adjust assumptions
Drag a slider; the chart updates instantly. Sliders override the selected worldview, and switching worldviews resets them.
How to read this
The everyday causes (heart disease, cancer, etc.) come from US government mortality records: the Social Security Administration's life table and CDC cause-of-death data.
The AIrows are estimates, not records. No mortality table exists for risks that (mostly) haven't happened yet, so the model breaks each one into smaller questions: when the capability arrives, how often things go wrong, how many people die. Estimated answers are from published forecasts, public statements, and expert surveys. Every input is adjustable in the panel under the chart, and every source is cited at the bottom of the page.
Why would AI be a cause of death?
Loss of control. AI companies are racing to build systems smarter than humans. Modern AI is grown by training, not written line by line, so no one can yet verify what goals end up inside it.[src] Geoffrey Hinton and Yoshua Bengio, two founders of the field, warn that a smarter-than-human AI could slip past human oversight and seize control of events.[src][src] The danger is that an AI steering the future redirects the energy, land, and materials people need to live.[src]
Engineered pandemics. AI is lowering the barrier to designing and building dangerous pathogens.[src] Expert forecasters already put the yearly odds of a serious AI-enabled epidemic several times above the pre-AI baseline.[src]
Weaponized AI. AI inside nuclear early-warning systems could push a crisis over the edge,[src] and autonomous weapons and AI-driven attacks on civilians or infrastructure could kill at scale.[src][src]
Take action
Lower these odds
The real outcome depends on choices made in the next few years. Pick how you want to help — contacting your representatives is extremely effective and only takes about three minutes.
Other organizations to support
Sources & method
This is an estimative model, not a prediction. The conventional-mortality numbers are real US data: the SSA Period Life Table 2022 and CDC WONDER 2018–2024 cause-of-death fractions averaged between sexes, good enough for a rough estimate, though other countries have different patterns. The AI-risk numbers are conditional estimates broken out of the cited literature and expert forecasts. Every input is adjustable above and sourced here.