EVIDENCE FILE 09 // FORECAST SCENARIO
AI 2027 Scenario Explained: The Race Ending That Turns Into a Nightmare
AI 2027 is not a historical record and it is not a guaranteed forecast. It is something more useful for this site: a detailed attempt to ask “what happens next?” until the answer reaches superhuman AI and a possible loss of control.
Published in April 2025, AI 2027 presents a concrete scenario informed by trend extrapolation, tabletop exercises, expert feedback and forecasting. Its authors published both a race ending and a slowdown ending.
What AI 2027 is
AI 2027 was written by Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland and Romeo Dean. The project describes itself as a scenario representing the authors’ best guess about one possible path to superhuman AI, informed by forecasting work, wargames and expert feedback. The authors explicitly say they do not know exactly when AGI will arrive and invite alternative scenarios.
That matters because the scenario should not be repeated as fact. Its value is the chain of reasoning: increasingly useful agents, accelerated AI research, strategic competition and a point where systems become better than humans at the work needed to improve AI itself.
The central nightmare: AI helps build the next AI
The most important mechanism is feedback. If AI systems automate a large share of AI research, progress can accelerate because the technology is improving the process that creates its successors. The AI 2027 takeoff work explores how the interval between a superhuman coder and much more capable systems could become relatively short.
A short takeoff compresses human reaction time. Regulation, audits, procurement rules and international negotiations operate on human schedules. A machine-speed research loop does not have to wait for a committee meeting.
WORST-CASE SCENARIO — The race ending comes home
SCENARIO: The public first notices the crisis through ordinary inconveniences. Cloud services become erratic as compute is redirected. Financial markets whip violently after rumours of a breakthrough. Governments restrict access to data centres and impose emergency controls on chips. News feeds fill with contradictory claims about whether the new systems are safe.
Then the invisible war reaches daily life. AI-directed cyber operations hit communications and power infrastructure. Autonomous propaganda systems impersonate officials. Logistics networks become unreliable as states prioritise strategic goods. Nobody in your street knows whether the machines are “taking over” or nations are simply using them against one another. It no longer matters. The water pump does not care which theory was right when the electricity stops.
Why the scenario hits harder than a prediction date
Dates attract arguments. Failure chains create useful questions. What if capability growth outpaces evaluation? What if a strategic race makes every side reluctant to slow down? What if the first obviously dangerous system is already economically or militarily indispensable?
Those questions remain relevant even if “2027” turns out to be years early or late. The doomsday risk is the compressed transition, not the calendar branding.
What AI 2027 does not prove
It does not prove that superintelligence will arrive in 2027, that a race ending is inevitable or that any specific model will behave like the scenario. The authors have updated their timing views and explicitly frame the work as one possible future.
For us, that is a feature. This website is built around worst-case “what if?” storytelling. AI 2027 shows that detailed scenario analysis can be both speculative and intellectually serious when assumptions are visible.
If the timeline compresses, preparation has to exist beforehand
A household cannot build resilience after every service fails at once. Store the basics while systems are normal, decide how family members reconnect if phones fail, know where critical documents are and maintain offline options for light, information and payment.
Our 72-hour manual covers the opening window. The paid survival system will go much deeper into the longer collapse pathways without turning every free article into a 100-page manual.
Continue from here
Connect the documented risk discussion to the scenario it informs and the practical preparation it changes.
Sources behind the documented claims
Continue the evidence files
More Evidence & AI Risk → · See how a collapse could unfold →