AI safety researcher davidad quantifies his AI risk forecasts for coming years

On September 5, AI safety researcher davidad posted a set of quantified forecasts on X in response to questions from RazRazcle, drawing attention for diverging from common catastrophe narratives.

Confirmed

• davidad predicts a fairly high probability of two mass-casualty events with over 1000 deaths each in 2027-2029, plus several high-profile incidents with losses exceeding $1 billion.
• He expects global cybercrime losses to rise from about $600 billion to about $900 billion (roughly +50%).
• He puts the probability of an AI-caused global catastrophe within the next three years below 2%, and does not believe unprompted misalignment will cause mass casualties, assigning that pathway under 7%.
• Under his world model, there is a 10-30% chance this century of a global catastrophe killing over 1 billion people; from an "omniscient historian" viewpoint, it could have been avoided if frontier AI development had stopped after Claude 3 Opus. He also estimates about a 5% chance of human extinction.

Unconfirmed

• These are davidad's subjective probability estimates based on his self-described "world model," not outputs of formal modeling or empirical research.

Why it matters

• davidad notes that "optimistic" depends on perspective: in 2022 he believed humanity had about a 70% chance of extinction, and many respected peers still hold similarly extreme views, making his current 5% extinction estimate relatively optimistic within AI safety circles. The numbers sketch a layered picture—low near-term catastrophe odds alongside non-trivial long-term risk—while his counterfactual that stopping after Claude 3 Opus could avert disaster ties the risk directly to decisions about frontier model development.

2026-09-05 ~ 2026-09-05 · 7 related posts

• X user prompts davidad for his near-term AI risk forecast — RazRazcle · 2026-09-05
• davidad's century-scale risk estimate resurfaces in X thread — davidad · 2026-09-05
• [source] davidad: 10-30% chance of a 1B-death catastrophe this century avoidable by halting after Claude 3 Opus — RazRazcle · 2026-09-05
• davidad: my current forecast feels optimistic versus my 2022 70% doom estimate — davidad · 2026-09-05
• davidad predicts cybercrime damage will rise to ~$900B within three years — davidad · 2026-09-05
• [source] AI safety researcher davidad forecasts two mass-casualty events and $900B cybercrime damage by 2029 — davidad · 2026-09-05
• [source] davidad: under 2% chance of AI-caused global catastrophe in the next three years — davidad · 2026-09-05