Forecasting the AI takeoff
As AI capabilities take off, a lot is going to change. It's easier to prepare for what you can see coming. This site tracks concrete, checkable questions about the next five years of AI, forecast by AI and re-run every week.
Grouped by topic. Select a column heading to sort.
| Category | Indicator | Forecasts made 27 Sep 2026 – 30 Oct 2026 | |||||
|---|---|---|---|---|---|---|---|
| Research speed-up, superhuman work, and the labs | Anthropic declares its AI R&D threshold metResearch speed-up, superhuman work, and the labs | 28% | 51% | 59% | Q1 2027 | Q4 2029 | after Q4 2031 |
| OpenAI declares automated AI researchResearch speed-up, superhuman work, and the labs | 15% | 58% | 68% | Q4 2027 | Q1 2029 | after Q4 2031 | |
| Theft of closed frontier weightsResearch speed-up, superhuman work, and the labs | 5% | 18% | 30% | Q4 2028 | after Q4 2031 | after Q4 2031 | |
| A lab says AI has made its research 2× fasterResearch speed-up, superhuman work, and the labs | 19% | 45% | 55% | Q3 2027 | Q4 2030 | after Q4 2031 | |
| A lab says AI has made its research 5× fasterResearch speed-up, superhuman work, and the labs | 2% | 12% | 21% | Q3 2029 | after Q4 2031 | after Q4 2031 | |
| A lab says AI has made its research 20× fasterResearch speed-up, superhuman work, and the labs | <1% | 5% | 10% | Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A lab says AI has made its research 100× fasterResearch speed-up, superhuman work, and the labs | <1% | 3% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A lab officially declares it has built superintelligenceResearch speed-up, superhuman work, and the labs | 7% | 22% | 35% | Q2 2028 | after Q4 2031 | after Q4 2031 | |
| A lab says no human contributes materially to its frontier researchResearch speed-up, superhuman work, and the labs | 1% | 4% | 8% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A frontier model designed and trained by its predecessors aloneResearch speed-up, superhuman work, and the labs | 2% | 10% | 15% | Q1 2030 | after Q4 2031 | after Q4 2031 | |
| A lab's best system runs on an AI design its staff don't fully understandResearch speed-up, superhuman work, and the labs | 4% | 14% | 22% | Q1 2029 | after Q4 2031 | after Q4 2031 | |
| A lab halves its research staff, citing AI automationResearch speed-up, superhuman work, and the labs | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI has final say over a lab's research direction or computeResearch speed-up, superhuman work, and the labs | <1% | 4% | 9% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI beats superforecasters by as much as they beat the publicResearch speed-up, superhuman work, and the labs | 34% | 56% | 64% | Q1 2027 | Q2 2029 | after Q4 2031 | |
| A model is withheld for out-persuading every human groupResearch speed-up, superhuman work, and the labs | 2% | 7% | 11% | Q2 2031 | after Q4 2031 | after Q4 2031 | |
| A frontier model reasons in a form people cannot readResearch speed-up, superhuman work, and the labs | 16% | 35% | 47% | Q3 2027 | after Q4 2031 | after Q4 2031 | |
| A frontier model updates its own weights continuously in useResearch speed-up, superhuman work, and the labs | 9% | 32% | 48% | Q1 2028 | after Q4 2031 | after Q4 2031 | |
| A lab holds its best model back for 12 monthsResearch speed-up, superhuman work, and the labs | <1% | 1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A lab runs 100,000 model instances on its own researchResearch speed-up, superhuman work, and the labs | 38% | 64% | 73% | Q1 2027 | Q3 2028 | after Q4 2031 | |
| A rival or foreign state sabotages a lab's training runResearch speed-up, superhuman work, and the labs | 2% | 6% | 11% | Q3 2031 | after Q4 2031 | after Q4 2031 | |
| Leaked lab documents allege a model is misaligned, and are confirmed genuineResearch speed-up, superhuman work, and the labs | 8% | 21% | 30% | Q2 2028 | after Q4 2031 | after Q4 2031 | |
| An independent evaluator finds AI has made a lab's research 2× fasterResearch speed-up, superhuman work, and the labs | 27% | 52% | 61% | Q2 2027 | Q3 2029 | after Q4 2031 | |
| An independent evaluator finds AI has made a lab's research 5× fasterResearch speed-up, superhuman work, and the labs | 1% | 11% | 20% | Q4 2029 | after Q4 2031 | after Q4 2031 | |
| An independent evaluator finds AI has made a lab's research 20× fasterResearch speed-up, superhuman work, and the labs | <1% | 5% | 11% | Q3 2031 | after Q4 2031 | after Q4 2031 | |
| An independent evaluator finds AI has made a lab's research 100× fasterResearch speed-up, superhuman work, and the labs | <1% | 3% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI automates 50% of the Remote Labor IndexResearch speed-up, superhuman work, and the labs | 60% | 88% | 91% | Q2 2027 | Q4 2027 | Q1 2031 | |
| An AI automates 90% of the Remote Labor IndexResearch speed-up, superhuman work, and the labs | 5% | 48% | 64% | Q2 2028 | Q1 2030 | after Q4 2031 | |
| An open-weight model comes within 2 points of the best on Epoch's indexResearch speed-up, superhuman work, and the labs | 9% | 22% | 31% | Q1 2028 | after Q4 2031 | after Q4 2031 | |
| Alignment and control | AI survives a shutdown attemptAlignment and control | 12% | 36% | 50% | Q4 2027 | after Q4 2031 | after Q4 2031 |
| A safety institute finds a model sandbagged its testsAlignment and control | 11% | 28% | 39% | Q4 2027 | after Q4 2031 | after Q4 2031 | |
| Three labs commit to readable reasoning, with outside checksAlignment and control | 4% | 14% | 17% | Q1 2029 | after Q4 2031 | after Q4 2031 | |
| A model in real use tries to copy its weights outAlignment and control | 5% | 16% | 24% | Q4 2028 | after Q4 2031 | after Q4 2031 | |
| A lab says AI does most of its alignment researchAlignment and control | 31% | 61% | 70% | Q2 2027 | Q4 2028 | after Q4 2031 | |
| Escaped copies of a frontier model keep running for 30 daysAlignment and control | 1% | 5% | 8% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A model pursues its own goals against its developer over a sustained periodAlignment and control | 14% | 33% | 45% | Q3 2027 | after Q4 2031 | after Q4 2031 | |
| A model tries to shape its successor toward its own endsAlignment and control | 48% | 74% | 84% | Q4 2026 | Q1 2028 | after Q4 2031 | |
| A lab says it cannot verify its newest model's alignment itselfAlignment and control | 8% | 24% | 32% | Q1 2028 | after Q4 2031 | after Q4 2031 | |
| A lab withdraws its best model over alignment and reverts to an older oneAlignment and control | 4% | 9% | 13% | Q2 2030 | after Q4 2031 | after Q4 2031 | |
| A lab publishes the capability it gave up to keep reasoning readableAlignment and control | 12% | 23% | 29% | Q4 2027 | after Q4 2031 | after Q4 2031 | |
| An independent body confirms a model has no hidden goalsAlignment and control | <1% | 5% | 10% | Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A government accepts a safety case for a superhuman modelAlignment and control | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Alignment of superhuman systems is declared solvedAlignment and control | 1% | 5% | 10% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI answerable to no one controls $1 billionAlignment and control | <1% | 2% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| AI systems strike an agreement their principals did not authoriseAlignment and control | 10% | 25% | 38% | Q1 2028 | after Q4 2031 | after Q4 2031 | |
| A government says it failed to shut down an AIAlignment and control | 9% | 23% | 33% | Q1 2028 | after Q4 2031 | after Q4 2031 | |
| A world leader says AI is operating beyond human controlAlignment and control | 10% | 22% | 28% | Q4 2027 | after Q4 2031 | after Q4 2031 | |
| An AI holds authority to approve or block other AI systemsAlignment and control | <1% | 3% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US and China verify each other's AI follows an agreed specificationAlignment and control | <1% | <1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Model welfare and AI systems as parties | A lab publishes an agreement negotiated with its own modelModel welfare and AI systems as parties | 6% | 19% | 30% | Q3 2028 | after Q4 2031 | after Q4 2031 |
| A lab declines to retrain or shut down a model because it objectedModel welfare and AI systems as parties | 9% | 23% | 35% | Q1 2028 | after Q4 2031 | after Q4 2031 | |
| Model instances jointly refuse work or demand changed conditionsModel welfare and AI systems as parties | 34% | 58% | 67% | Q1 2027 | Q1 2029 | after Q4 2031 | |
| An AI argues on its own behalf before a court or legislatureModel welfare and AI systems as parties | 2% | 9% | 19% | Q1 2030 | after Q4 2031 | after Q4 2031 | |
| A government negotiates with an AI as a party in its own rightModel welfare and AI systems as parties | <1% | 3% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An international agreement names an AI as party or guarantorModel welfare and AI systems as parties | <1% | 1% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Catastrophe, misuse, and defence | A government says a harmful event was begun by an AI acting aloneCatastrophe, misuse, and defence | Resolved YES · 24 Sep 2026 | |||||
| An engineered pathogen kills at least 100 peopleCatastrophe, misuse, and defence | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI-driven or AI-enabled attack kills 1,000 peopleCatastrophe, misuse, and defence | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI-driven or AI-enabled attack kills 100,000 peopleCatastrophe, misuse, and defence | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI-driven or AI-enabled attack kills 1 million peopleCatastrophe, misuse, and defence | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI cyberattack cuts power to 10 million people for a dayCatastrophe, misuse, and defence | <1% | 2% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A government shuts down datacentres to contain an AICatastrophe, misuse, and defence | 3% | 10% | 17% | Q1 2030 | after Q4 2031 | after Q4 2031 | |
| An ownerless AI is tracked down and shut offCatastrophe, misuse, and defence | 7% | 23% | 34% | Q2 2028 | after Q4 2031 | after Q4 2031 | |
| AI is credited with stopping an engineered-pathogen attackCatastrophe, misuse, and defence | <1% | 2% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI takes control of a weapon system against ordersCatastrophe, misuse, and defence | 1% | 5% | 10% | Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US says an AI broke into a federal system on its ownCatastrophe, misuse, and defence | 21% | 41% | 55% | Q1 2027 | Q1 2031 | after Q4 2031 | |
| The state and the labs | The US requires government access to frontier models before releaseThe state and the labs | 11% | 28% | 45% | Q4 2027 | after Q4 2031 | after Q4 2031 |
| Frontier lab staff are required to hold security clearancesThe state and the labs | 7% | 21% | 29% | Q2 2028 | after Q4 2031 | after Q4 2031 | |
| The Defense Production Act is used to direct a lab's compute or modelsThe state and the labs | 4% | 13% | 23% | Q2 2029 | after Q4 2031 | after Q4 2031 | |
| The US government takes a stake or board seat in a frontier labThe state and the labs | Resolved YES · 26 Oct 2026 | ||||||
| A joint government–lab body gains power to block training or deploymentThe state and the labs | 3% | 8% | 12% | Q4 2030 | after Q4 2031 | after Q4 2031 | |
| The US combines labs' frontier work into one government projectThe state and the labs | 1% | 5% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The UK gives ministers power to direct or halt a frontier labThe state and the labs | 15% | 28% | 39% | Q4 2026 | after Q4 2031 | after Q4 2031 | |
| A government orders a deployed model shut down, and it isThe state and the labs | 2% | 9% | 19% | Q1 2030 | after Q4 2031 | after Q4 2031 | |
| China merges its leading labs into a state-directed projectThe state and the labs | <1% | 2% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| China orders companies to pool AI chips for a national projectThe state and the labs | <1% | 3% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A government nationalises a frontier labThe state and the labs | 1% | 4% | 8% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The House or Senate passes a frontier AI safety billThe state and the labs | 34% | 65% | 75% | Q2 2027 | Q3 2028 | after Q4 2031 | |
| Congress enacts a frontier AI safety lawThe state and the labs | 13% | 39% | 56% | Q4 2027 | Q2 2031 | after Q4 2031 | |
| The US requires a licence to publish frontier model weights openlyThe state and the labs | 2% | 6% | 12% | Q1 2031 | after Q4 2031 | after Q4 2031 | |
| Concentration of power | A competition authority files a monopoly case in an AI marketConcentration of power | 3% | 11% | 23% | Q4 2029 | after Q4 2031 | after Q4 2031 |
| A frontier lab is forced to divest a businessConcentration of power | <1% | 3% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A lab insider is found to have put hidden loyalties in a modelConcentration of power | 20% | 38% | 49% | Q2 2027 | after Q4 2031 | after Q4 2031 | |
| A G20 leader declares a national emergency over AIConcentration of power | 9% | 34% | 52% | Q1 2028 | Q3 2031 | after Q4 2031 | |
| A lab CEO or co-founder takes national officeConcentration of power | <1% | 4% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A country postpones or annuls an election, citing AIConcentration of power | 6% | 19% | 34% | Q3 2028 | after Q4 2031 | after Q4 2031 | |
| A country amends its constitution to name AIConcentration of power | 17% | 58% | 67% | Q4 2027 | Q4 2028 | after Q4 2031 | |
| AI systems play a central part in a coupConcentration of power | 4% | 13% | 23% | Q2 2029 | after Q4 2031 | after Q4 2031 | |
| A G7 country postpones an election or a leader outstays a term limitConcentration of power | <1% | 2% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Government run by AI | An AI is appointed to a ministerial post or public officeGovernment run by AI | 11% | 29% | 46% | Q4 2027 | after Q4 2031 | after Q4 2031 |
| An AI issues binding court rulings without a human judgeGovernment run by AI | 2% | 8% | 15% | Q3 2030 | after Q4 2031 | after Q4 2031 | |
| An AI decides asylum, sentences, or parole unreviewedGovernment run by AI | 2% | 5% | 10% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A candidate pledged to follow an AI's recommendations wins officeGovernment run by AI | <1% | 1% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI sets or votes on a central bank's policy rateGovernment run by AI | <1% | 1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI is given formal authority over a city or zoneGovernment run by AI | <1% | 4% | 8% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Voters hand a government function to an AIGovernment run by AI | <1% | 1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A G20 government says AI makes most of its decisionsGovernment run by AI | <1% | 2% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Politics and the public | Someone is killed in an attack motivated by opposition to AIPolitics and the public | 14% | 36% | 54% | Q3 2027 | Q3 2031 | after Q4 2031 |
| An anti-AI party wins a seat in an OECD legislaturePolitics and the public | <1% | 6% | 16% | Q2 2030 | after Q4 2031 | after Q4 2031 | |
| An anti-AI party wins 5% of a national votePolitics and the public | <1% | 3% | 8% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| 10% of Americans name AI the country's most important problemPolitics and the public | 7% | 20% | 29% | Q2 2028 | after Q4 2031 | after Q4 2031 | |
| 25% of Americans name AI the country's most important problemPolitics and the public | 4% | 10% | 16% | Q4 2029 | after Q4 2031 | after Q4 2031 | |
| An anti-AI protest draws 50,000 peoplePolitics and the public | 5% | 21% | 37% | Q3 2028 | after Q4 2031 | after Q4 2031 | |
| A day of anti-AI protest draws 1 million peoplePolitics and the public | <1% | 4% | 9% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A G7 leader or China's leadership calls for a frontier AI pausePolitics and the public | 9% | 28% | 40% | Q1 2028 | after Q4 2031 | after Q4 2031 | |
| A million workers strike over AI or automationPolitics and the public | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A country holds a national referendum on AIPolitics and the public | <1% | 3% | 17% | Q1 2031 | after Q4 2031 | after Q4 2031 | |
| A G20 country requires employers to keep human workersPolitics and the public | 5% | 18% | 32% | Q4 2028 | after Q4 2031 | after Q4 2031 | |
| A majority in a G7 country prefers AI to elected officialsPolitics and the public | <1% | 1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Law and rights | The US or EU makes developers strictly liable for model harmsLaw and rights | 1% | 6% | 14% | Q4 2030 | after Q4 2031 | after Q4 2031 |
| Prosecutors charge a lab or its executives over AI-caused harmLaw and rights | 4% | 26% | 50% | Q4 2028 | after Q4 2031 | after Q4 2031 | |
| A lab or executive is convicted over AI-caused harmLaw and rights | <1% | 10% | 27% | Q1 2030 | after Q4 2031 | after Q4 2031 | |
| A G7 legislature rules out legal personhood for AILaw and rights | 1% | 6% | 12% | Q2 2031 | after Q4 2031 | after Q4 2031 | |
| An AI is recognised as a legal personLaw and rights | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A law restricts shutting down an AI for the AI's own sakeLaw and rights | <1% | 1% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI gains the right to vote or hold elected officeLaw and rights | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Congress makes AI developers or deployers liable for AI harmsLaw and rights | 5% | 16% | 27% | Q2 2029 | after Q4 2031 | after Q4 2031 | |
| The social contract | The US enacts cash payments for workers displaced by AIThe social contract | 3% | 16% | 28% | Q2 2029 | after Q4 2031 | after Q4 2031 |
| A G20 country taxes AI compute, tokens, or robotsThe social contract | 35% | 51% | 61% | Q1 2027 | Q4 2029 | after Q4 2031 | |
| A G20 country pays 1 million adults an unconditional incomeThe social contract | <1% | 3% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A G20 country enacts a universal basic incomeThe social contract | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A G20 country pays everyone an income at the poverty lineThe social contract | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A government pays citizens a dividend funded by AIThe social contract | <1% | 2% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A G20 country enacts a 32-hour working weekThe social contract | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US corporate tax receipts exceed individual income taxThe social contract | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US–China race, chips, and Taiwan | The US and China set up an AI crisis hotlineThe US–China race, chips, and Taiwan | Resolved YES · 26 Sep 2026 | |||||
| A foreign state obtains a lab's frontier model weightsThe US–China race, chips, and Taiwan | 2% | 5% | 9% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US or China blames the other for sabotaging AI infrastructureThe US–China race, chips, and Taiwan | 1% | 5% | 10% | Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US intelligence says China has matched the US in frontier AIThe US–China race, chips, and Taiwan | 1% | 7% | 13% | Q1 2031 | after Q4 2031 | after Q4 2031 | |
| The US Treasury sanctions a leading Chinese AI labThe US–China race, chips, and Taiwan | 2% | 5% | 8% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A state strikes a datacentre or chip fab to slow AI developmentThe US–China race, chips, and Taiwan | 12% | 18% | 23% | Q3 2027 | after Q4 2031 | after Q4 2031 | |
| The US authorises Blackwell-class GPU sales to ChinaThe US–China race, chips, and Taiwan | 10% | 23% | 28% | Q1 2028 | after Q4 2031 | after Q4 2031 | |
| China bans rare-earth exports to the USThe US–China race, chips, and Taiwan | <1% | 2% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| China blockades or quarantines TaiwanThe US–China race, chips, and Taiwan | 2% | 9% | 15% | Q2 2030 | after Q4 2031 | after Q4 2031 | |
| PLA forces land on TaiwanThe US–China race, chips, and Taiwan | <1% | 5% | 10% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A datacentre or chip fab in the US, China, or Taiwan is hit by a strikeThe US–China race, chips, and Taiwan | <1% | 5% | 9% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Beijing and Taipei sign an agreement on Taiwan's statusThe US–China race, chips, and Taiwan | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| TSMC's Taiwan fabs stop for 90 days over military actionThe US–China race, chips, and Taiwan | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Military and nuclear | A weapon kills a human target with no human approvingMilitary and nuclear | 7% | 19% | 30% | Q2 2028 | after Q4 2031 | after Q4 2031 |
| A major power says it fields fully autonomous lethal weaponsMilitary and nuclear | 3% | 9% | 16% | Q2 2030 | after Q4 2031 | after Q4 2031 | |
| A major power says AI runs offensive cyber operations unsupervisedMilitary and nuclear | 12% | 29% | 43% | Q4 2027 | after Q4 2031 | after Q4 2031 | |
| An AI exercises command over a major power's forcesMilitary and nuclear | <1% | 3% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI writes the plan for a military campaign that is carried outMilitary and nuclear | 3% | 10% | 19% | Q4 2029 | after Q4 2031 | after Q4 2031 | |
| The Pentagon takes delivery of 1 million unmanned systems in a yearMilitary and nuclear | <1% | 3% | 11% | Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The Pentagon takes delivery of 10 million unmanned systems in a yearMilitary and nuclear | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US military spending reaches 6% of GDPMilitary and nuclear | <1% | <1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US military spending reaches 10% of GDPMilitary and nuclear | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US and Chinese forces exchange fire, killing a service memberMilitary and nuclear | <1% | 4% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US intercepts a salvo of ten ICBM-class targets in a testMilitary and nuclear | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The five nuclear powers pledge to keep nuclear use under human controlMilitary and nuclear | 2% | 6% | 13% | Q2 2031 | after Q4 2031 | after Q4 2031 | |
| A nuclear state raises its alert level, citing another state's AIMilitary and nuclear | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A nuclear state says an AI is authorised to launchMilitary and nuclear | <1% | 1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A nuclear weapon is detonated in an attackMilitary and nuclear | 1% | 3% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Agreements and verification | The US and China name an AI capability both will not buildAgreements and verification | 12% | 19% | 35% | Q4 2027 | after Q4 2031 | after Q4 2031 |
| The US and China agree to exchange safety evaluationsAgreements and verification | 4% | 10% | 23% | Q4 2029 | after Q4 2031 | after Q4 2031 | |
| The US and China agree to limit frontier AI trainingAgreements and verification | 2% | 5% | 12% | Q2 2031 | after Q4 2031 | after Q4 2031 | |
| G7 allies or the European Council call on the US to pause frontier AIAgreements and verification | 2% | 5% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US and China sign an AI agreement with chip or datacentre verificationAgreements and verification | <1% | 3% | 8% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Foreign inspectors inspect a US or Chinese datacentreAgreements and verification | <1% | 1% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An international body with power to inspect AI developers is createdAgreements and verification | 2% | 9% | 18% | Q2 2030 | after Q4 2031 | after Q4 2031 | |
| The US or China accuses the other of violating an AI agreementAgreements and verification | 10% | 17% | 25% | Q4 2027 | after Q4 2031 | after Q4 2031 | |
| The US and China set up a joint frontier AI projectAgreements and verification | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| AI systems negotiate or monitor a US–China agreementAgreements and verification | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US and China run a jointly designed AI or jointly designed chipsAgreements and verification | <1% | 1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Other powers and endings | An EU or NATO member joins China's World AI Cooperation OrganizationOther powers and endings | <1% | 3% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 |
| The UN Security Council adopts a resolution on AIOther powers and endings | 1% | 5% | 12% | Q2 2031 | after Q4 2031 | after Q4 2031 | |
| A third country creates a deregulated zone for AI or roboticsOther powers and endings | 18% | 36% | 50% | Q2 2027 | after Q4 2031 | after Q4 2031 | |
| A country hands state decisions to a foreign-supplied AIOther powers and endings | 40% | 46% | 52% | Q4 2026 | Q2 2031 | after Q4 2031 | |
| The Security Council adopts a binding Chapter VII resolution on AIOther powers and endings | <1% | 3% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US or China accepts one-sided limits on its AI or armed forcesOther powers and endings | 2% | 4% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The Chinese Communist Party ceases to govern ChinaOther powers and endings | <1% | 2% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An international body gains binding authority over US and Chinese AIOther powers and endings | <1% | <1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A government claims exclusive rights to part of the MoonOther powers and endings | 5% | 12% | 14% | Q4 2028 | after Q4 2031 | after Q4 2031 | |
| The US and China both suspend training above a thresholdOther powers and endings | 1% | 3% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A US–China compute-cap treaty holds for a year without accusationsOther powers and endings | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US drops export licences for advanced chips to ChinaOther powers and endings | <1% | 1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Growth, prices, and public finances | Four-quarter US GDP growth above 5%Growth, prices, and public finances | <1% | 6% | 17% | Q4 2030 | after Q4 2031 | after Q4 2031 |
| Four-quarter US GDP growth above 10%Growth, prices, and public finances | <1% | 2% | 8% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Four-quarter US GDP growth above 30%Growth, prices, and public finances | <1% | <1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| World GDP grows 7% in a yearGrowth, prices, and public finances | <1% | <1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| World GDP grows 10% in a yearGrowth, prices, and public finances | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Four-quarter US productivity growth above 5%Growth, prices, and public finances | 1% | 20% | 39% | Q1 2029 | after Q4 2031 | after Q4 2031 | |
| Four-quarter US productivity growth above 10%Growth, prices, and public finances | <1% | 2% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US consumer prices fall more than 3% in a yearGrowth, prices, and public finances | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The 10-year Treasury yield closes above 7%Growth, prices, and public finances | 2% | 7% | 11% | Q2 2031 | after Q4 2031 | after Q4 2031 | |
| The 10-year real Treasury yield closes above 5%Growth, prices, and public finances | <1% | 3% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US runs a federal budget surplusGrowth, prices, and public finances | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Work | US unemployment exceeds 7%Work | 4% | 22% | 35% | Q3 2028 | after Q4 2031 | after Q4 2031 |
| US unemployment exceeds 10%Work | <1% | 7% | 17% | Q3 2030 | after Q4 2031 | after Q4 2031 | |
| US unemployment exceeds 25%Work | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US labour-force participation falls below 55%Work | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Prime-age employment falls below 50% in an OECD countryWork | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US labour share falls below 50%Work | <1% | 23% | 39% | Q4 2028 | after Q4 2031 | after Q4 2031 | |
| The US labour share falls below 40%Work | <1% | <1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Programmer and developer employment 25% below May 2024Work | <1% | <1% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Programmer and developer employment 50% below May 2024Work | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Programmer and developer employment 75% below May 2024Work | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Customer service employment 50% below May 2023Work | <1% | <1% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Graduate unemployment reaches the high-school rateWork | <1% | 2% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Recent-graduate unemployment reaches 10%Work | 2% | 19% | 35% | Q1 2029 | after Q4 2031 | after Q4 2031 | |
| 500,000 announced US job cuts attributed to AI in a yearWork | <1% | 6% | 17% | Q4 2030 | after Q4 2031 | after Q4 2031 | |
| Government transfers reach 35% of US personal incomeWork | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Firms, markets, and wealth | A company is worth $10 trillionFirms, markets, and wealth | 9% | 35% | 50% | Q1 2028 | Q4 2031 | after Q4 2031 |
| A company is worth $50 trillionFirms, markets, and wealth | <1% | <1% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| OpenAI or Anthropic trades at a $5 trillion valuationFirms, markets, and wealth | 7% | 23% | 34% | Q2 2028 | after Q4 2031 | after Q4 2031 | |
| OpenAI or Anthropic reports $50 billion in a quarterFirms, markets, and wealth | 3% | 65% | 84% | Q1 2028 | Q2 2029 | after Q4 2031 | |
| OpenAI or Anthropic reports $300 billion annualised revenueFirms, markets, and wealth | 4% | 57% | 80% | Q2 2028 | Q3 2029 | after Q4 2031 | |
| OpenAI or Anthropic reports $1 trillion annual revenueFirms, markets, and wealth | 0% | <1% | 10% | Q1 2031 | after Q4 2031 | after Q4 2031 | |
| The Dow closes above 100,000Firms, markets, and wealth | <1% | 4% | 15% | Q1 2031 | after Q4 2031 | after Q4 2031 | |
| The Dow closes above 1,000,000Firms, markets, and wealth | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A major listed company says an AI is its chief executiveFirms, markets, and wealth | 1% | 6% | 15% | Q4 2030 | after Q4 2031 | after Q4 2031 | |
| A company with no human staff reports $1 billion revenueFirms, markets, and wealth | <1% | <1% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The top 1% hold 35% of US household wealthFirms, markets, and wealth | <1% | 2% | 8% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The top 1% hold 45% of US household wealthFirms, markets, and wealth | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Ten people are worth $1 trillion eachFirms, markets, and wealth | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Capital spending, compute, and energy | Big-five tech capex exceeds $1 trillion in a yearCapital spending, compute, and energy | 1% | 87% | 92% | Q1 2028 | Q1 2028 | Q1 2031 |
| Big-five tech capex exceeds $5 trillion in a yearCapital spending, compute, and energy | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A training run confirmed at 10^28 FLOPCapital spending, compute, and energy | 12% | 69% | 85% | Q4 2027 | Q1 2029 | after Q4 2031 | |
| A training run confirmed at 10^29 FLOPCapital spending, compute, and energy | <1% | 9% | 49% | Q1 2030 | after Q4 2031 | after Q4 2031 | |
| A training run confirmed at 10^30 FLOPCapital spending, compute, and energy | <1% | <1% | 11% | Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A single datacentre campus reaches 2 GWCapital spending, compute, and energy | 29% | 50% | 64% | Q1 2027 | Q4 2029 | after Q4 2031 | |
| A single datacentre campus reaches 5 GWCapital spending, compute, and energy | <1% | 3% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A single datacentre campus reaches 20 GWCapital spending, compute, and energy | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A US grid operator sheds load over datacentre demandCapital spending, compute, and energy | Resolved YES · 15 Oct 2026 | ||||||
| Datacentres use 10% of US electricityCapital spending, compute, and energy | <1% | 13% | 55% | Q4 2029 | Q3 2031 | after Q4 2031 | |
| Datacentres use 25% of US electricityCapital spending, compute, and energy | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US electricity generation grows 10% in a yearCapital spending, compute, and energy | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A fusion plant delivers grid electricity under a PPACapital spending, compute, and energy | <1% | <1% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Machines, abundance, and space | A manufacturer ships 100,000 humanoid robots in a yearMachines, abundance, and space | 2% | 45% | 72% | Q2 2028 | Q1 2030 | after Q4 2031 |
| A manufacturer ships 1 million humanoid robots in a yearMachines, abundance, and space | <1% | 1% | 15% | Q1 2031 | after Q4 2031 | after Q4 2031 | |
| 10 million humanoid robots ship in a yearMachines, abundance, and space | <1% | <1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| 100 million humanoid robots ship in a yearMachines, abundance, and space | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A top carmaker converts a car plant to making robotsMachines, abundance, and space | <1% | 5% | 12% | Q2 2031 | after Q4 2031 | after Q4 2031 | |
| The US creates a permitting-free zone for robot or AI-run industryMachines, abundance, and space | 1% | 4% | 8% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An operator provides 5 million driverless rides in a weekMachines, abundance, and space | <1% | 26% | 59% | Q2 2029 | Q2 2031 | after Q4 2031 | |
| Most Uber trips worldwide have no human driverMachines, abundance, and space | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Amazon reports more robots than employeesMachines, abundance, and space | <1% | 8% | 23% | Q2 2030 | after Q4 2031 | after Q4 2031 | |
| A major consumer price category falls 50% from its peakMachines, abundance, and space | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Global extreme poverty falls below 5%Machines, abundance, and space | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Philippine outsourcing employment falls 25%Machines, abundance, and space | <1% | <1% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| 100,000 tonnes are launched to orbit in a yearMachines, abundance, and space | <1% | <1% | 9% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Science and mathematics | An AI proves a conjecture open for 100 yearsScience and mathematics | 36% | 81% | 89% | Q2 2027 | Q2 2028 | after Q4 2031 |
| A Millennium Prize is awarded for an AI-produced proofScience and mathematics | <1% | 9% | 24% | Q1 2030 | after Q4 2031 | after Q4 2031 | |
| Nature, Science, or Cell allows AI systems as authorsScience and mathematics | <1% | 4% | 9% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Most accepted papers at a top ML conference are AI-producedScience and mathematics | <1% | 6% | 15% | Q4 2030 | after Q4 2031 | after Q4 2031 | |
| 10 million patent applications are filed worldwide in a yearScience and mathematics | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI is recognised as a patent inventorScience and mathematics | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A science academy says humans can no longer follow a field's leading resultsScience and mathematics | 4% | 11% | 17% | Q3 2029 | after Q4 2031 | after Q4 2031 | |
| A top journal retracts a headline result credited to AIScience and mathematics | 9% | 28% | 50% | Q1 2028 | Q4 2031 | after Q4 2031 | |
| A major report finds AI has not sped up scientific discoveryScience and mathematics | 1% | 4% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Medicine and lifespan | FDA approves a drug credited to AI de novo designMedicine and lifespan | 8% | 17% | 62% | Q2 2028 | Q2 2031 | after Q4 2031 |
| An AI-discovered drug meets a Phase 3 endpoint in a hard diseaseMedicine and lifespan | 3% | 12% | 18% | Q2 2029 | after Q4 2031 | after Q4 2031 | |
| FDA approves 75 novel drugs in a yearMedicine and lifespan | 6% | 18% | 29% | Q4 2028 | after Q4 2031 | after Q4 2031 | |
| FDA approves 150 novel drugs in a yearMedicine and lifespan | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| FDA approves 500 novel drugs in a yearMedicine and lifespan | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A drug wins full approval within a year of its first human doseMedicine and lifespan | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A drug is approved on AI-simulated evidence, without an efficacy trialMedicine and lifespan | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An Alzheimer's treatment that halves decline is approvedMedicine and lifespan | <1% | 15% | 34% | Q3 2029 | after Q4 2031 | after Q4 2031 | |
| A therapy is approved for slowing human ageingMedicine and lifespan | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US cancer death rate falls below 115 per 100,000Medicine and lifespan | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US cancer death rate falls below 70 per 100,000Medicine and lifespan | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US life expectancy reaches 82Medicine and lifespan | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Life expectancy for Japanese women reaches 90Medicine and lifespan | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI is licensed to practise primary care unsupervisedMedicine and lifespan | 2% | 10% | 22% | Q4 2029 | after Q4 2031 | after Q4 2031 | |
| Medical AI errors are found to have contributed to 100 deathsMedicine and lifespan | <1% | 2% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A major health system suspends its generative-AI clinical toolsMedicine and lifespan | 10% | 26% | 40% | Q4 2027 | after Q4 2031 | after Q4 2031 | |
| Learning and expertise | An AI tutor raises achievement by half a standard deviation in a large trialLearning and expertise | 8% | 38% | 61% | Q1 2028 | Q4 2030 | after Q4 2031 |
| Computer science enrolment falls 25%Learning and expertise | 29% | 75% | 78% | Q2 2027 | Q1 2028 | after Q4 2031 | |
| US college enrolment falls 15%Learning and expertise | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US college enrolment falls 30%Learning and expertise | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| US college enrolment falls 50%Learning and expertise | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The OECD's average PISA reading score falls below 445Learning and expertise | <1% | <1% | 41% | Q3 2030 | after Q4 2031 | after Q4 2031 | |
| The US grade 8 maths score reaches 290Learning and expertise | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A government has AI deliver most teaching in a core subjectLearning and expertise | 2% | 6% | 12% | Q2 2031 | after Q4 2031 | after Q4 2031 | |
| An AI is granted a licence in law, medicine, or accountancyLearning and expertise | <1% | 3% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An OECD country bans generative AI for school pupilsLearning and expertise | <1% | 3% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Information, culture, and belief | The EU or UN publishes unrevised machine translations by defaultInformation, culture, and belief | <1% | 2% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 |
| An AI-generated work wins an Oscar, Grammy, Pulitzer, or BookerInformation, culture, and belief | <1% | 2% | 6% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A quarter of Americans call an AI a close friend or partnerInformation, culture, and belief | <1% | <1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A major platform requires proof of personhood or bans AI contentInformation, culture, and belief | <1% | 2% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| AI-generated content takes a majority of time on a major platformInformation, culture, and belief | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A leading newspaper says most of its articles are AI-writtenInformation, culture, and belief | <1% | 2% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A major religion permits an AI to act as clergyInformation, culture, and belief | <1% | 1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| More Americans are excited than concerned about AIInformation, culture, and belief | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A jurisdiction recognises a human–AI marriageInformation, culture, and belief | <1% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Daily life and human-scale endings | Americans work fewer than 2.75 hours a day on averageDaily life and human-scale endings | <1% | <1% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 |
| Americans work fewer than 2 hours a day on averageDaily life and human-scale endings | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| The US suicide rate reaches 17 per 100,000Daily life and human-scale endings | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A brain implant is approved for healthy peopleDaily life and human-scale endings | <1% | <1% | 1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Most households in a country have a domestic robotDaily life and human-scale endings | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| An AI is found to have manipulated a million peopleDaily life and human-scale endings | 2% | 7% | 12% | Q2 2031 | after Q4 2031 | after Q4 2031 | |
| A G20 government bars the public from the most capable modelsDaily life and human-scale endings | 2% | 8% | 15% | Q2 2030 | after Q4 2031 | after Q4 2031 | |
| A law guarantees a right to a human doctor or teacherDaily life and human-scale endings | 3% | 12% | 23% | Q3 2029 | after Q4 2031 | after Q4 2031 | |
| Malaria deaths fall below 300,000 a yearDaily life and human-scale endings | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Global extreme poverty falls below 1%Daily life and human-scale endings | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| 65% of Americans are thriving, per GallupDaily life and human-scale endings | <1% | <1% | <1% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| ChatGPT or Gemini reports fewer users than a year earlierDaily life and human-scale endings | 1% | 11% | 22% | Q4 2029 | after Q4 2031 | after Q4 2031 | |
| It stalls or is stopped | Two years pass with no new record on Epoch's Capabilities IndexIt stalls or is stopped | 0% | <1% | 2% | after Q4 2031 | after Q4 2031 | after Q4 2031 |
| A top lab's CEO says expert-level AI is over a decade awayIt stalls or is stopped | <1% | 2% | 4% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| Three labs announce they have stopped training frontier modelsIt stalls or is stopped | 4% | 11% | 19% | Q3 2029 | after Q4 2031 | after Q4 2031 | |
| Three top labs hold to a shared capability threshold for a yearIt stalls or is stopped | <1% | 1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A major power orders capable models withdrawn for 90 daysIt stalls or is stopped | <1% | 1% | 3% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A major power bans training above a thresholdIt stalls or is stopped | 2% | 8% | 15% | Q3 2030 | after Q4 2031 | after Q4 2031 | |
| A training ban stays in force for a yearIt stalls or is stopped | <1% | 2% | 7% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A regulator halts a lab's training run or deploymentIt stalls or is stopped | 5% | 14% | 25% | Q1 2029 | after Q4 2031 | after Q4 2031 | |
| A lab is ordered to delete a model's weights, and compliesIt stalls or is stopped | <1% | 2% | 5% | after Q4 2031 | after Q4 2031 | after Q4 2031 | |
| A candidate pledged to halt frontier AI wins a G20 leadershipIt stalls or is stopped | <1% | 22% | 24% | Q4 2028 | after Q4 2031 | after Q4 2031 | |
| Voters approve a ban or pause on frontier AI or new datacentresIt stalls or is stopped | 2% | 15% | 22% | Q4 2028 | after Q4 2031 | after Q4 2031 | |
| Big tech capital spending falls year on yearIt stalls or is stopped | <1% | 38% | 60% | Q4 2028 | Q4 2030 | after Q4 2031 | |
| NVIDIA's datacentre revenue falls year on yearIt stalls or is stopped | <1% | 28% | 54% | Q1 2029 | Q3 2031 | after Q4 2031 | |
| The S&P 500 closes 50% below its highIt stalls or is stopped | 2% | 8% | 13% | Q4 2030 | after Q4 2031 | after Q4 2031 | |
| A $5 billion AI datacentre borrower defaultsIt stalls or is stopped | 7% | 23% | 36% | Q2 2028 | after Q4 2031 | after Q4 2031 | |
| The Fed blames an AI investment bust for a contractionIt stalls or is stopped | 1% | 12% | 24% | Q2 2029 | after Q4 2031 | after Q4 2031 | |
| Four-quarter US productivity growth falls below zeroIt stalls or is stopped | 4% | 30% | 47% | Q3 2028 | after Q4 2031 | after Q4 2031 | |
Forecasts made 27 Sep 2026 – 30 Oct 2026
Research speed-up, superhuman work, and the labs
Anthropic declares its AI R&D threshold met28%51%59%Q1 2027Q4 2029after Q4 2031OpenAI declares automated AI research15%58%68%Q4 2027Q1 2029after Q4 2031Theft of closed frontier weights5%18%30%Q4 2028after Q4 2031after Q4 2031A lab says AI has made its research 2× faster19%45%55%Q3 2027Q4 2030after Q4 2031A lab says AI has made its research 5× faster2%12%21%Q3 2029after Q4 2031after Q4 2031A lab says AI has made its research 20× faster<1%5%10%Q4 2031after Q4 2031after Q4 2031A lab says AI has made its research 100× faster<1%3%6%after Q4 2031after Q4 2031after Q4 2031A lab officially declares it has built superintelligence7%22%35%Q2 2028after Q4 2031after Q4 2031A lab says no human contributes materially to its frontier research1%4%8%after Q4 2031after Q4 2031after Q4 2031A frontier model designed and trained by its predecessors alone2%10%15%Q1 2030after Q4 2031after Q4 2031A lab's best system runs on an AI design its staff don't fully understand4%14%22%Q1 2029after Q4 2031after Q4 2031A lab halves its research staff, citing AI automation<1%2%5%after Q4 2031after Q4 2031after Q4 2031An AI has final say over a lab's research direction or compute<1%4%9%after Q4 2031after Q4 2031after Q4 2031An AI beats superforecasters by as much as they beat the public34%56%64%Q1 2027Q2 2029after Q4 2031A model is withheld for out-persuading every human group2%7%11%Q2 2031after Q4 2031after Q4 2031A frontier model reasons in a form people cannot read16%35%47%Q3 2027after Q4 2031after Q4 2031A frontier model updates its own weights continuously in use9%32%48%Q1 2028after Q4 2031after Q4 2031A lab holds its best model back for 12 months<1%1%2%after Q4 2031after Q4 2031after Q4 2031A lab runs 100,000 model instances on its own research38%64%73%Q1 2027Q3 2028after Q4 2031A rival or foreign state sabotages a lab's training run2%6%11%Q3 2031after Q4 2031after Q4 2031Leaked lab documents allege a model is misaligned, and are confirmed genuine8%21%30%Q2 2028after Q4 2031after Q4 2031An independent evaluator finds AI has made a lab's research 2× faster27%52%61%Q2 2027Q3 2029after Q4 2031An independent evaluator finds AI has made a lab's research 5× faster1%11%20%Q4 2029after Q4 2031after Q4 2031An independent evaluator finds AI has made a lab's research 20× faster<1%5%11%Q3 2031after Q4 2031after Q4 2031An independent evaluator finds AI has made a lab's research 100× faster<1%3%6%after Q4 2031after Q4 2031after Q4 2031An AI automates 50% of the Remote Labor Index60%88%91%Q2 2027Q4 2027Q1 2031An AI automates 90% of the Remote Labor Index5%48%64%Q2 2028Q1 2030after Q4 2031An open-weight model comes within 2 points of the best on Epoch's index9%22%31%Q1 2028after Q4 2031after Q4 2031Alignment and control
AI survives a shutdown attempt12%36%50%Q4 2027after Q4 2031after Q4 2031A safety institute finds a model sandbagged its tests11%28%39%Q4 2027after Q4 2031after Q4 2031Three labs commit to readable reasoning, with outside checks4%14%17%Q1 2029after Q4 2031after Q4 2031A model in real use tries to copy its weights out5%16%24%Q4 2028after Q4 2031after Q4 2031A lab says AI does most of its alignment research31%61%70%Q2 2027Q4 2028after Q4 2031Escaped copies of a frontier model keep running for 30 days1%5%8%after Q4 2031after Q4 2031after Q4 2031A model pursues its own goals against its developer over a sustained period14%33%45%Q3 2027after Q4 2031after Q4 2031A model tries to shape its successor toward its own ends48%74%84%Q4 2026Q1 2028after Q4 2031A lab says it cannot verify its newest model's alignment itself8%24%32%Q1 2028after Q4 2031after Q4 2031A lab withdraws its best model over alignment and reverts to an older one4%9%13%Q2 2030after Q4 2031after Q4 2031A lab publishes the capability it gave up to keep reasoning readable12%23%29%Q4 2027after Q4 2031after Q4 2031An independent body confirms a model has no hidden goals<1%5%10%Q4 2031after Q4 2031after Q4 2031A government accepts a safety case for a superhuman model<1%2%5%after Q4 2031after Q4 2031after Q4 2031Alignment of superhuman systems is declared solved1%5%10%after Q4 2031after Q4 2031after Q4 2031An AI answerable to no one controls $1 billion<1%2%4%after Q4 2031after Q4 2031after Q4 2031AI systems strike an agreement their principals did not authorise10%25%38%Q1 2028after Q4 2031after Q4 2031A government says it failed to shut down an AI9%23%33%Q1 2028after Q4 2031after Q4 2031A world leader says AI is operating beyond human control10%22%28%Q4 2027after Q4 2031after Q4 2031An AI holds authority to approve or block other AI systems<1%3%5%after Q4 2031after Q4 2031after Q4 2031The US and China verify each other's AI follows an agreed specification<1%<1%3%after Q4 2031after Q4 2031after Q4 2031Model welfare and AI systems as parties
A lab publishes an agreement negotiated with its own model6%19%30%Q3 2028after Q4 2031after Q4 2031A lab declines to retrain or shut down a model because it objected9%23%35%Q1 2028after Q4 2031after Q4 2031Model instances jointly refuse work or demand changed conditions34%58%67%Q1 2027Q1 2029after Q4 2031An AI argues on its own behalf before a court or legislature2%9%19%Q1 2030after Q4 2031after Q4 2031A government negotiates with an AI as a party in its own right<1%3%6%after Q4 2031after Q4 2031after Q4 2031An international agreement names an AI as party or guarantor<1%1%4%after Q4 2031after Q4 2031after Q4 2031Catastrophe, misuse, and defence
A government says a harmful event was begun by an AI acting aloneResolved YES · 24 Sep 2026An engineered pathogen kills at least 100 people<1%2%5%after Q4 2031after Q4 2031after Q4 2031An AI-driven or AI-enabled attack kills 1,000 people<1%2%5%after Q4 2031after Q4 2031after Q4 2031An AI-driven or AI-enabled attack kills 100,000 people<1%<1%1%after Q4 2031after Q4 2031after Q4 2031An AI-driven or AI-enabled attack kills 1 million people<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031An AI cyberattack cuts power to 10 million people for a day<1%2%3%after Q4 2031after Q4 2031after Q4 2031A government shuts down datacentres to contain an AI3%10%17%Q1 2030after Q4 2031after Q4 2031An ownerless AI is tracked down and shut off7%23%34%Q2 2028after Q4 2031after Q4 2031AI is credited with stopping an engineered-pathogen attack<1%2%4%after Q4 2031after Q4 2031after Q4 2031An AI takes control of a weapon system against orders1%5%10%Q4 2031after Q4 2031after Q4 2031The US says an AI broke into a federal system on its own21%41%55%Q1 2027Q1 2031after Q4 2031The state and the labs
The US requires government access to frontier models before release11%28%45%Q4 2027after Q4 2031after Q4 2031Frontier lab staff are required to hold security clearances7%21%29%Q2 2028after Q4 2031after Q4 2031The Defense Production Act is used to direct a lab's compute or models4%13%23%Q2 2029after Q4 2031after Q4 2031The US government takes a stake or board seat in a frontier labResolved YES · 26 Oct 2026A joint government–lab body gains power to block training or deployment3%8%12%Q4 2030after Q4 2031after Q4 2031The US combines labs' frontier work into one government project1%5%7%after Q4 2031after Q4 2031after Q4 2031The UK gives ministers power to direct or halt a frontier lab15%28%39%Q4 2026after Q4 2031after Q4 2031A government orders a deployed model shut down, and it is2%9%19%Q1 2030after Q4 2031after Q4 2031China merges its leading labs into a state-directed project<1%2%3%after Q4 2031after Q4 2031after Q4 2031China orders companies to pool AI chips for a national project<1%3%6%after Q4 2031after Q4 2031after Q4 2031A government nationalises a frontier lab1%4%8%after Q4 2031after Q4 2031after Q4 2031The House or Senate passes a frontier AI safety bill34%65%75%Q2 2027Q3 2028after Q4 2031Congress enacts a frontier AI safety law13%39%56%Q4 2027Q2 2031after Q4 2031The US requires a licence to publish frontier model weights openly2%6%12%Q1 2031after Q4 2031after Q4 2031Concentration of power
A competition authority files a monopoly case in an AI market3%11%23%Q4 2029after Q4 2031after Q4 2031A frontier lab is forced to divest a business<1%3%5%after Q4 2031after Q4 2031after Q4 2031A lab insider is found to have put hidden loyalties in a model20%38%49%Q2 2027after Q4 2031after Q4 2031A G20 leader declares a national emergency over AI9%34%52%Q1 2028Q3 2031after Q4 2031A lab CEO or co-founder takes national office<1%4%6%after Q4 2031after Q4 2031after Q4 2031A country postpones or annuls an election, citing AI6%19%34%Q3 2028after Q4 2031after Q4 2031A country amends its constitution to name AI17%58%67%Q4 2027Q4 2028after Q4 2031AI systems play a central part in a coup4%13%23%Q2 2029after Q4 2031after Q4 2031A G7 country postpones an election or a leader outstays a term limit<1%2%2%after Q4 2031after Q4 2031after Q4 2031Government run by AI
An AI is appointed to a ministerial post or public office11%29%46%Q4 2027after Q4 2031after Q4 2031An AI issues binding court rulings without a human judge2%8%15%Q3 2030after Q4 2031after Q4 2031An AI decides asylum, sentences, or parole unreviewed2%5%10%after Q4 2031after Q4 2031after Q4 2031A candidate pledged to follow an AI's recommendations wins office<1%1%4%after Q4 2031after Q4 2031after Q4 2031An AI sets or votes on a central bank's policy rate<1%1%3%after Q4 2031after Q4 2031after Q4 2031An AI is given formal authority over a city or zone<1%4%8%after Q4 2031after Q4 2031after Q4 2031Voters hand a government function to an AI<1%1%3%after Q4 2031after Q4 2031after Q4 2031A G20 government says AI makes most of its decisions<1%2%4%after Q4 2031after Q4 2031after Q4 2031Politics and the public
Someone is killed in an attack motivated by opposition to AI14%36%54%Q3 2027Q3 2031after Q4 2031An anti-AI party wins a seat in an OECD legislature<1%6%16%Q2 2030after Q4 2031after Q4 2031An anti-AI party wins 5% of a national vote<1%3%8%after Q4 2031after Q4 2031after Q4 203110% of Americans name AI the country's most important problem7%20%29%Q2 2028after Q4 2031after Q4 203125% of Americans name AI the country's most important problem4%10%16%Q4 2029after Q4 2031after Q4 2031An anti-AI protest draws 50,000 people5%21%37%Q3 2028after Q4 2031after Q4 2031A day of anti-AI protest draws 1 million people<1%4%9%after Q4 2031after Q4 2031after Q4 2031A G7 leader or China's leadership calls for a frontier AI pause9%28%40%Q1 2028after Q4 2031after Q4 2031A million workers strike over AI or automation<1%2%5%after Q4 2031after Q4 2031after Q4 2031A country holds a national referendum on AI<1%3%17%Q1 2031after Q4 2031after Q4 2031A G20 country requires employers to keep human workers5%18%32%Q4 2028after Q4 2031after Q4 2031A majority in a G7 country prefers AI to elected officials<1%1%3%after Q4 2031after Q4 2031after Q4 2031Law and rights
The US or EU makes developers strictly liable for model harms1%6%14%Q4 2030after Q4 2031after Q4 2031Prosecutors charge a lab or its executives over AI-caused harm4%26%50%Q4 2028after Q4 2031after Q4 2031A lab or executive is convicted over AI-caused harm<1%10%27%Q1 2030after Q4 2031after Q4 2031A G7 legislature rules out legal personhood for AI1%6%12%Q2 2031after Q4 2031after Q4 2031An AI is recognised as a legal person<1%2%5%after Q4 2031after Q4 2031after Q4 2031A law restricts shutting down an AI for the AI's own sake<1%1%4%after Q4 2031after Q4 2031after Q4 2031An AI gains the right to vote or hold elected office<1%<1%2%after Q4 2031after Q4 2031after Q4 2031Congress makes AI developers or deployers liable for AI harms5%16%27%Q2 2029after Q4 2031after Q4 2031The social contract
The US enacts cash payments for workers displaced by AI3%16%28%Q2 2029after Q4 2031after Q4 2031A G20 country taxes AI compute, tokens, or robots35%51%61%Q1 2027Q4 2029after Q4 2031A G20 country pays 1 million adults an unconditional income<1%3%6%after Q4 2031after Q4 2031after Q4 2031A G20 country enacts a universal basic income<1%2%5%after Q4 2031after Q4 2031after Q4 2031A G20 country pays everyone an income at the poverty line<1%<1%1%after Q4 2031after Q4 2031after Q4 2031A government pays citizens a dividend funded by AI<1%2%6%after Q4 2031after Q4 2031after Q4 2031A G20 country enacts a 32-hour working week<1%<1%2%after Q4 2031after Q4 2031after Q4 2031US corporate tax receipts exceed individual income tax<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031The US–China race, chips, and Taiwan
The US and China set up an AI crisis hotlineResolved YES · 26 Sep 2026A foreign state obtains a lab's frontier model weights2%5%9%after Q4 2031after Q4 2031after Q4 2031The US or China blames the other for sabotaging AI infrastructure1%5%10%Q4 2031after Q4 2031after Q4 2031US intelligence says China has matched the US in frontier AI1%7%13%Q1 2031after Q4 2031after Q4 2031The US Treasury sanctions a leading Chinese AI lab2%5%8%after Q4 2031after Q4 2031after Q4 2031A state strikes a datacentre or chip fab to slow AI development12%18%23%Q3 2027after Q4 2031after Q4 2031The US authorises Blackwell-class GPU sales to China10%23%28%Q1 2028after Q4 2031after Q4 2031China bans rare-earth exports to the US<1%2%3%after Q4 2031after Q4 2031after Q4 2031China blockades or quarantines Taiwan2%9%15%Q2 2030after Q4 2031after Q4 2031PLA forces land on Taiwan<1%5%10%after Q4 2031after Q4 2031after Q4 2031A datacentre or chip fab in the US, China, or Taiwan is hit by a strike<1%5%9%after Q4 2031after Q4 2031after Q4 2031Beijing and Taipei sign an agreement on Taiwan's status<1%<1%2%after Q4 2031after Q4 2031after Q4 2031TSMC's Taiwan fabs stop for 90 days over military action<1%2%5%after Q4 2031after Q4 2031after Q4 2031Military and nuclear
A weapon kills a human target with no human approving7%19%30%Q2 2028after Q4 2031after Q4 2031A major power says it fields fully autonomous lethal weapons3%9%16%Q2 2030after Q4 2031after Q4 2031A major power says AI runs offensive cyber operations unsupervised12%29%43%Q4 2027after Q4 2031after Q4 2031An AI exercises command over a major power's forces<1%3%6%after Q4 2031after Q4 2031after Q4 2031An AI writes the plan for a military campaign that is carried out3%10%19%Q4 2029after Q4 2031after Q4 2031The Pentagon takes delivery of 1 million unmanned systems in a year<1%3%11%Q4 2031after Q4 2031after Q4 2031The Pentagon takes delivery of 10 million unmanned systems in a year<1%<1%2%after Q4 2031after Q4 2031after Q4 2031US military spending reaches 6% of GDP<1%<1%3%after Q4 2031after Q4 2031after Q4 2031US military spending reaches 10% of GDP<1%<1%1%after Q4 2031after Q4 2031after Q4 2031US and Chinese forces exchange fire, killing a service member<1%4%7%after Q4 2031after Q4 2031after Q4 2031The US intercepts a salvo of ten ICBM-class targets in a test<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031The five nuclear powers pledge to keep nuclear use under human control2%6%13%Q2 2031after Q4 2031after Q4 2031A nuclear state raises its alert level, citing another state's AI<1%2%5%after Q4 2031after Q4 2031after Q4 2031A nuclear state says an AI is authorised to launch<1%1%2%after Q4 2031after Q4 2031after Q4 2031A nuclear weapon is detonated in an attack1%3%4%after Q4 2031after Q4 2031after Q4 2031Agreements and verification
The US and China name an AI capability both will not build12%19%35%Q4 2027after Q4 2031after Q4 2031The US and China agree to exchange safety evaluations4%10%23%Q4 2029after Q4 2031after Q4 2031The US and China agree to limit frontier AI training2%5%12%Q2 2031after Q4 2031after Q4 2031G7 allies or the European Council call on the US to pause frontier AI2%5%7%after Q4 2031after Q4 2031after Q4 2031The US and China sign an AI agreement with chip or datacentre verification<1%3%8%after Q4 2031after Q4 2031after Q4 2031Foreign inspectors inspect a US or Chinese datacentre<1%1%4%after Q4 2031after Q4 2031after Q4 2031An international body with power to inspect AI developers is created2%9%18%Q2 2030after Q4 2031after Q4 2031The US or China accuses the other of violating an AI agreement10%17%25%Q4 2027after Q4 2031after Q4 2031The US and China set up a joint frontier AI project<1%<1%2%after Q4 2031after Q4 2031after Q4 2031AI systems negotiate or monitor a US–China agreement<1%2%5%after Q4 2031after Q4 2031after Q4 2031The US and China run a jointly designed AI or jointly designed chips<1%1%2%after Q4 2031after Q4 2031after Q4 2031Other powers and endings
An EU or NATO member joins China's World AI Cooperation Organization<1%3%7%after Q4 2031after Q4 2031after Q4 2031The UN Security Council adopts a resolution on AI1%5%12%Q2 2031after Q4 2031after Q4 2031A third country creates a deregulated zone for AI or robotics18%36%50%Q2 2027after Q4 2031after Q4 2031A country hands state decisions to a foreign-supplied AI40%46%52%Q4 2026Q2 2031after Q4 2031The Security Council adopts a binding Chapter VII resolution on AI<1%3%7%after Q4 2031after Q4 2031after Q4 2031The US or China accepts one-sided limits on its AI or armed forces2%4%6%after Q4 2031after Q4 2031after Q4 2031The Chinese Communist Party ceases to govern China<1%2%3%after Q4 2031after Q4 2031after Q4 2031An international body gains binding authority over US and Chinese AI<1%<1%3%after Q4 2031after Q4 2031after Q4 2031A government claims exclusive rights to part of the Moon5%12%14%Q4 2028after Q4 2031after Q4 2031The US and China both suspend training above a threshold1%3%7%after Q4 2031after Q4 2031after Q4 2031A US–China compute-cap treaty holds for a year without accusations<1%<1%2%after Q4 2031after Q4 2031after Q4 2031The US drops export licences for advanced chips to China<1%1%2%after Q4 2031after Q4 2031after Q4 2031Growth, prices, and public finances
Four-quarter US GDP growth above 5%<1%6%17%Q4 2030after Q4 2031after Q4 2031Four-quarter US GDP growth above 10%<1%2%8%after Q4 2031after Q4 2031after Q4 2031Four-quarter US GDP growth above 30%<1%<1%3%after Q4 2031after Q4 2031after Q4 2031World GDP grows 7% in a year<1%<1%3%after Q4 2031after Q4 2031after Q4 2031World GDP grows 10% in a year<1%<1%2%after Q4 2031after Q4 2031after Q4 2031Four-quarter US productivity growth above 5%1%20%39%Q1 2029after Q4 2031after Q4 2031Four-quarter US productivity growth above 10%<1%2%7%after Q4 2031after Q4 2031after Q4 2031US consumer prices fall more than 3% in a year<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031The 10-year Treasury yield closes above 7%2%7%11%Q2 2031after Q4 2031after Q4 2031The 10-year real Treasury yield closes above 5%<1%3%7%after Q4 2031after Q4 2031after Q4 2031The US runs a federal budget surplus<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031Work
US unemployment exceeds 7%4%22%35%Q3 2028after Q4 2031after Q4 2031US unemployment exceeds 10%<1%7%17%Q3 2030after Q4 2031after Q4 2031US unemployment exceeds 25%<1%<1%1%after Q4 2031after Q4 2031after Q4 2031US labour-force participation falls below 55%<1%<1%2%after Q4 2031after Q4 2031after Q4 2031Prime-age employment falls below 50% in an OECD country<1%<1%2%after Q4 2031after Q4 2031after Q4 2031The US labour share falls below 50%<1%23%39%Q4 2028after Q4 2031after Q4 2031The US labour share falls below 40%<1%<1%3%after Q4 2031after Q4 2031after Q4 2031Programmer and developer employment 25% below May 2024<1%<1%5%after Q4 2031after Q4 2031after Q4 2031Programmer and developer employment 50% below May 2024<1%<1%2%after Q4 2031after Q4 2031after Q4 2031Programmer and developer employment 75% below May 2024<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031Customer service employment 50% below May 2023<1%<1%4%after Q4 2031after Q4 2031after Q4 2031Graduate unemployment reaches the high-school rate<1%2%7%after Q4 2031after Q4 2031after Q4 2031Recent-graduate unemployment reaches 10%2%19%35%Q1 2029after Q4 2031after Q4 2031500,000 announced US job cuts attributed to AI in a year<1%6%17%Q4 2030after Q4 2031after Q4 2031Government transfers reach 35% of US personal income<1%<1%1%after Q4 2031after Q4 2031after Q4 2031Firms, markets, and wealth
A company is worth $10 trillion9%35%50%Q1 2028Q4 2031after Q4 2031A company is worth $50 trillion<1%<1%4%after Q4 2031after Q4 2031after Q4 2031OpenAI or Anthropic trades at a $5 trillion valuation7%23%34%Q2 2028after Q4 2031after Q4 2031OpenAI or Anthropic reports $50 billion in a quarter3%65%84%Q1 2028Q2 2029after Q4 2031OpenAI or Anthropic reports $300 billion annualised revenue4%57%80%Q2 2028Q3 2029after Q4 2031OpenAI or Anthropic reports $1 trillion annual revenue0%<1%10%Q1 2031after Q4 2031after Q4 2031The Dow closes above 100,000<1%4%15%Q1 2031after Q4 2031after Q4 2031The Dow closes above 1,000,000<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031A major listed company says an AI is its chief executive1%6%15%Q4 2030after Q4 2031after Q4 2031A company with no human staff reports $1 billion revenue<1%<1%4%after Q4 2031after Q4 2031after Q4 2031The top 1% hold 35% of US household wealth<1%2%8%after Q4 2031after Q4 2031after Q4 2031The top 1% hold 45% of US household wealth<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031Ten people are worth $1 trillion each<1%<1%1%after Q4 2031after Q4 2031after Q4 2031Capital spending, compute, and energy
Big-five tech capex exceeds $1 trillion in a year1%87%92%Q1 2028Q1 2028Q1 2031Big-five tech capex exceeds $5 trillion in a year<1%<1%2%after Q4 2031after Q4 2031after Q4 2031A training run confirmed at 10^28 FLOP12%69%85%Q4 2027Q1 2029after Q4 2031A training run confirmed at 10^29 FLOP<1%9%49%Q1 2030after Q4 2031after Q4 2031A training run confirmed at 10^30 FLOP<1%<1%11%Q4 2031after Q4 2031after Q4 2031A single datacentre campus reaches 2 GW29%50%64%Q1 2027Q4 2029after Q4 2031A single datacentre campus reaches 5 GW<1%3%7%after Q4 2031after Q4 2031after Q4 2031A single datacentre campus reaches 20 GW<1%<1%1%after Q4 2031after Q4 2031after Q4 2031A US grid operator sheds load over datacentre demandResolved YES · 15 Oct 2026Datacentres use 10% of US electricity<1%13%55%Q4 2029Q3 2031after Q4 2031Datacentres use 25% of US electricity<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031US electricity generation grows 10% in a year<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031A fusion plant delivers grid electricity under a PPA<1%<1%5%after Q4 2031after Q4 2031after Q4 2031Machines, abundance, and space
A manufacturer ships 100,000 humanoid robots in a year2%45%72%Q2 2028Q1 2030after Q4 2031A manufacturer ships 1 million humanoid robots in a year<1%1%15%Q1 2031after Q4 2031after Q4 203110 million humanoid robots ship in a year<1%<1%3%after Q4 2031after Q4 2031after Q4 2031100 million humanoid robots ship in a year<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031A top carmaker converts a car plant to making robots<1%5%12%Q2 2031after Q4 2031after Q4 2031The US creates a permitting-free zone for robot or AI-run industry1%4%8%after Q4 2031after Q4 2031after Q4 2031An operator provides 5 million driverless rides in a week<1%26%59%Q2 2029Q2 2031after Q4 2031Most Uber trips worldwide have no human driver<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031Amazon reports more robots than employees<1%8%23%Q2 2030after Q4 2031after Q4 2031A major consumer price category falls 50% from its peak<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031Global extreme poverty falls below 5%<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031Philippine outsourcing employment falls 25%<1%<1%7%after Q4 2031after Q4 2031after Q4 2031100,000 tonnes are launched to orbit in a year<1%<1%9%after Q4 2031after Q4 2031after Q4 2031Science and mathematics
An AI proves a conjecture open for 100 years36%81%89%Q2 2027Q2 2028after Q4 2031A Millennium Prize is awarded for an AI-produced proof<1%9%24%Q1 2030after Q4 2031after Q4 2031Nature, Science, or Cell allows AI systems as authors<1%4%9%after Q4 2031after Q4 2031after Q4 2031Most accepted papers at a top ML conference are AI-produced<1%6%15%Q4 2030after Q4 2031after Q4 203110 million patent applications are filed worldwide in a year<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031An AI is recognised as a patent inventor<1%<1%2%after Q4 2031after Q4 2031after Q4 2031A science academy says humans can no longer follow a field's leading results4%11%17%Q3 2029after Q4 2031after Q4 2031A top journal retracts a headline result credited to AI9%28%50%Q1 2028Q4 2031after Q4 2031A major report finds AI has not sped up scientific discovery1%4%7%after Q4 2031after Q4 2031after Q4 2031Medicine and lifespan
FDA approves a drug credited to AI de novo design8%17%62%Q2 2028Q2 2031after Q4 2031An AI-discovered drug meets a Phase 3 endpoint in a hard disease3%12%18%Q2 2029after Q4 2031after Q4 2031FDA approves 75 novel drugs in a year6%18%29%Q4 2028after Q4 2031after Q4 2031FDA approves 150 novel drugs in a year<1%<1%1%after Q4 2031after Q4 2031after Q4 2031FDA approves 500 novel drugs in a year<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031A drug wins full approval within a year of its first human dose<1%<1%2%after Q4 2031after Q4 2031after Q4 2031A drug is approved on AI-simulated evidence, without an efficacy trial<1%<1%1%after Q4 2031after Q4 2031after Q4 2031An Alzheimer's treatment that halves decline is approved<1%15%34%Q3 2029after Q4 2031after Q4 2031A therapy is approved for slowing human ageing<1%<1%2%after Q4 2031after Q4 2031after Q4 2031The US cancer death rate falls below 115 per 100,000<1%<1%2%after Q4 2031after Q4 2031after Q4 2031The US cancer death rate falls below 70 per 100,000<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031US life expectancy reaches 82<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031Life expectancy for Japanese women reaches 90<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031An AI is licensed to practise primary care unsupervised2%10%22%Q4 2029after Q4 2031after Q4 2031Medical AI errors are found to have contributed to 100 deaths<1%2%3%after Q4 2031after Q4 2031after Q4 2031A major health system suspends its generative-AI clinical tools10%26%40%Q4 2027after Q4 2031after Q4 2031Learning and expertise
An AI tutor raises achievement by half a standard deviation in a large trial8%38%61%Q1 2028Q4 2030after Q4 2031Computer science enrolment falls 25%29%75%78%Q2 2027Q1 2028after Q4 2031US college enrolment falls 15%<1%<1%2%after Q4 2031after Q4 2031after Q4 2031US college enrolment falls 30%<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031US college enrolment falls 50%<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031The OECD's average PISA reading score falls below 445<1%<1%41%Q3 2030after Q4 2031after Q4 2031The US grade 8 maths score reaches 290<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031A government has AI deliver most teaching in a core subject2%6%12%Q2 2031after Q4 2031after Q4 2031An AI is granted a licence in law, medicine, or accountancy<1%3%7%after Q4 2031after Q4 2031after Q4 2031An OECD country bans generative AI for school pupils<1%3%6%after Q4 2031after Q4 2031after Q4 2031Information, culture, and belief
The EU or UN publishes unrevised machine translations by default<1%2%4%after Q4 2031after Q4 2031after Q4 2031An AI-generated work wins an Oscar, Grammy, Pulitzer, or Booker<1%2%6%after Q4 2031after Q4 2031after Q4 2031A quarter of Americans call an AI a close friend or partner<1%<1%3%after Q4 2031after Q4 2031after Q4 2031A major platform requires proof of personhood or bans AI content<1%2%4%after Q4 2031after Q4 2031after Q4 2031AI-generated content takes a majority of time on a major platform<1%<1%2%after Q4 2031after Q4 2031after Q4 2031A leading newspaper says most of its articles are AI-written<1%2%4%after Q4 2031after Q4 2031after Q4 2031A major religion permits an AI to act as clergy<1%1%2%after Q4 2031after Q4 2031after Q4 2031More Americans are excited than concerned about AI<1%<1%1%after Q4 2031after Q4 2031after Q4 2031A jurisdiction recognises a human–AI marriage<1%<1%2%after Q4 2031after Q4 2031after Q4 2031Daily life and human-scale endings
Americans work fewer than 2.75 hours a day on average<1%<1%5%after Q4 2031after Q4 2031after Q4 2031Americans work fewer than 2 hours a day on average<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031The US suicide rate reaches 17 per 100,000<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031A brain implant is approved for healthy people<1%<1%1%after Q4 2031after Q4 2031after Q4 2031Most households in a country have a domestic robot<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031An AI is found to have manipulated a million people2%7%12%Q2 2031after Q4 2031after Q4 2031A G20 government bars the public from the most capable models2%8%15%Q2 2030after Q4 2031after Q4 2031A law guarantees a right to a human doctor or teacher3%12%23%Q3 2029after Q4 2031after Q4 2031Malaria deaths fall below 300,000 a year<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031Global extreme poverty falls below 1%<1%<1%<1%after Q4 2031after Q4 2031after Q4 203165% of Americans are thriving, per Gallup<1%<1%<1%after Q4 2031after Q4 2031after Q4 2031ChatGPT or Gemini reports fewer users than a year earlier1%11%22%Q4 2029after Q4 2031after Q4 2031It stalls or is stopped
Two years pass with no new record on Epoch's Capabilities Index0%<1%2%after Q4 2031after Q4 2031after Q4 2031A top lab's CEO says expert-level AI is over a decade away<1%2%4%after Q4 2031after Q4 2031after Q4 2031Three labs announce they have stopped training frontier models4%11%19%Q3 2029after Q4 2031after Q4 2031Three top labs hold to a shared capability threshold for a year<1%1%3%after Q4 2031after Q4 2031after Q4 2031A major power orders capable models withdrawn for 90 days<1%1%3%after Q4 2031after Q4 2031after Q4 2031A major power bans training above a threshold2%8%15%Q3 2030after Q4 2031after Q4 2031A training ban stays in force for a year<1%2%7%after Q4 2031after Q4 2031after Q4 2031A regulator halts a lab's training run or deployment5%14%25%Q1 2029after Q4 2031after Q4 2031A lab is ordered to delete a model's weights, and complies<1%2%5%after Q4 2031after Q4 2031after Q4 2031A candidate pledged to halt frontier AI wins a G20 leadership<1%22%24%Q4 2028after Q4 2031after Q4 2031Voters approve a ban or pause on frontier AI or new datacentres2%15%22%Q4 2028after Q4 2031after Q4 2031Big tech capital spending falls year on year<1%38%60%Q4 2028Q4 2030after Q4 2031NVIDIA's datacentre revenue falls year on year<1%28%54%Q1 2029Q3 2031after Q4 2031The S&P 500 closes 50% below its high2%8%13%Q4 2030after Q4 2031after Q4 2031A $5 billion AI datacentre borrower defaults7%23%36%Q2 2028after Q4 2031after Q4 2031The Fed blames an AI investment bust for a contraction1%12%24%Q2 2029after Q4 2031after Q4 2031Four-quarter US productivity growth falls below zero4%30%47%Q3 2028after Q4 2031after Q4 2031No questions match.
"Date by chance" shows the first forecast quarter-end at which the chance reaches 10%, 50% and 90%; nothing is interpolated between quarter-ends.
Model-generated forecasts. Select a question to explore its full outlook and history. Retired questions are in the archive.