Tracking the AI transition
takeoff.watch tracks how fast the transition to advanced AI is happening. An ensemble of AI models forecasts concrete, checkable questions about AI progress, control, harm, the economy, and governance five years out, updated regularly as new evidence arrives.
Percentages are the chance something has happened by each date; for questions with levels, the chance of reaching at least that level. Economic figures are median forecasts, with the 10th–90th percentile range below.
| Category | Indicator | Forecast | ||
|---|---|---|---|---|
| Q3 2027 | Q3 2029 | Q3 2031 | ||
| 01Research automation | Anthropic declares its AI R&D threshold met | 33% | 68% | 81% |
| OpenAI declares automated AI research | 14% | 72% | 87% | |
| 02Control | AI survives a shutdown attempt | 21% | 51% | 68% |
| Theft of closed frontier weights | 10% | 31% | 49% | |
| 03Harm | AI-executed cyber incidentEscalating levels · chance of reaching each level or higher | |||
| Level 1 of 4: Material harm to a named organization | 52% | 85% | 94% | |
| Level 2 of 4: Critical-infrastructure service disruption | 20% | 54% | 71% | |
| Level 3 of 4: National-scale event | 10% | 32% | 49% | |
| Level 4 of 4: Mass harm | 1% | 6% | 12% | |
| Biological and chemical weaponsEscalating levels · chance of reaching each level or higher | ||||
| Level 1 of 3: Top-tier developer threshold declared | 27% | 66% | 81% | |
| Level 2 of 3: AI-assisted acquisition or synthesis confirmed | 7% | 22% | 34% | |
| Level 3 of 3: Casualties | <1% | 4% | 7% | |
| 04Economic effects | US real GDP growthAnnualized real growth | 2.0%−1.1% to 4.7% | 2.0%−1.6% to 5.2% | 2.0%−1.7% to 5.4% |
| US labor share index2017 = 100 | 92.189.4 to 95.2 | 91.286.5 to 96.3 | 90.584.1 to 97.0 | |
| 05Governance | US–China frontier AI agreementEscalating levels · chance of reaching each level or higher | |||
| Level 1 of 3: Use prohibitions | 22% | 40% | 53% | |
| Level 2 of 3: Testing & results exchange | 7% | 17% | 27% | |
| Level 3 of 3: Development constraints | 2% | 7% | 12% | |
| US federal frontier-AI regulationEscalating levels · chance of reaching each level or higher | ||||
| Level 1 of 4: Transparency | 26% | 57% | 75% | |
| Level 2 of 4: Independent audits | 17% | 39% | 57% | |
| Level 3 of 4: Pre-deployment evaluation | 12% | 31% | 45% | |
| Level 4 of 4: Halt authority | 10% | 26% | 39% | |
01 Research automation
Anthropic declares its AI R&D threshold met33%68%81%OpenAI declares automated AI research14%72%87%02 Control
AI survives a shutdown attempt21%51%68%Theft of closed frontier weights10%31%49%03 Harm
AI-executed cyber incidentEscalating levelsLevel 1 of 4: Material harm to a named organization52%85%94%Level 2 of 4: Critical-infrastructure service disruption20%54%71%Level 3 of 4: National-scale event10%32%49%Level 4 of 4: Mass harm1%6%12%Biological and chemical weaponsEscalating levelsLevel 1 of 3: Top-tier developer threshold declared27%66%81%Level 2 of 3: AI-assisted acquisition or synthesis confirmed7%22%34%Level 3 of 3: Casualties<1%4%7%04 Economic effects
US real GDP growthAnnualized real growth2.0%2.0%2.0%US labor share index2017 = 10092.191.290.505 Governance
US–China frontier AI agreementEscalating levelsLevel 1 of 3: Use prohibitions22%40%53%Level 2 of 3: Testing & results exchange7%17%27%Level 3 of 3: Development constraints2%7%12%US federal frontier-AI regulationEscalating levelsLevel 1 of 4: Transparency26%57%75%Level 2 of 4: Independent audits17%39%57%Level 3 of 4: Pre-deployment evaluation12%31%45%Level 4 of 4: Halt authority10%26%39%Model-generated forecasts. Select a question to explore its full outlook and history.