Sep 2028
Reliable Agent
Current Policy
In Discussion
Off The Table
62%
Employment
$50K
Median Income
2K
Researchers
0 mo
Total
Slodown
May 2030
Reliable Agent
Current Policy
In Discussion
Off The Table
44%
Employment
$52K
Median Income
4K
Researchers
1 yr
Total
Slodown
2033
Automated coder
Policy
Off The Table
13%
Employment
$10M
Median Income
50K
Researchers
3 yrs
Total
Slodown
2038
Top-Expert-Dominating AI
Policy
Off The Table
12%
Employment
$13M
Median Income
200K
Researchers
8 yrs
Total
Slodown
2040
Top-Expert-Dominating AI
This is a blueprint
from some of the
on how to survive
superintelligence.
And it says that America
data centers
inside Mongolia,
and China
its own
inside Canada.
what's going on.
AI Futures Project
We are a small research group forecasting the future of AI, funded by charitable
donations and grants. Our research team is Daniel Kokotajlo, Scott Alexander, Thomas Lennon, Romeo
Dean, Lauren Mungle, Milo Rodchen, and Nicole Serna. Explore our interactive
AI scenarios below. The distribution below are our team's probability on when
AI will achieve human-level coding performance.
AI
AI
from the
AI Futures Project,
the same
whose earlier
AI 2027
Daniel Kokotajlo, Scott Alexander, Thomas Larsen, Eli Lifland, Romeo Dean
We predict that the impact of superhuman AI over the next decade will be
enormous, exceeding that of the Industrial Revolution.
We wrote a scenario that represents our best guess about what that might look
like. It's informed by trend extrapolation, wargames, expert feedback, experience
at OpenAI, and previous forecasting successes.
What is this?
How did we write it?
Why is it valuable?
Who are we?
The CEOs of OpenAI, Google DeepMind, and Anthropic have all predicted that AGI will arrive
within the next 5 years. Sam Altman has said OpenAI is setting its sights on "superintelligence
in the true sense of the word" and the "glorious future."
What might that look like? We wrote AI 2027 to answer that question. Claims about the future
are often frustratingly vague, so we tried to be as concrete and quantitative as possible, even
though this means depicting one of many possible futures.
We wrote two endings: a "slowdown" and a "race" ending. However, AI 2027 is not a
recommendation or exhortation. Our goal is predictive accuracy.
We encourage you to debate and counter this scenario. We hope to spark a broad
conversation about where we're headed toward positive futures. We're
planning to give out thousands in prizes to the best alternative scenarios.
[Added Nov 22 2025, to prevent misunderstandings, we don't know exactly when AGI will be
built. 2027 was our medal (most likely) year at the time of publication, our medians were
somewhat longer. For our latest forecasts, see here.]
AI 2027
racing to
superhuman AI
in human extinction
or permanent
global dictatorship.
Plan A
Buy Time Slow down whenever is needed to have high confidence in safety.
Total Research Transparency Make almost all AI research fully visible to the public.
Diffuse AI Broadly Many companies in many countries at the frontier.
Reversibility Limit algorithmic progress; maintain Mutually Assured Compute Destruction.
Their new report,
lays out
how to
avoid both.
Mutually Assured Compute
Destruction
There is so, so much compute. Back in 2026-when many people thought AI was
an unsustainable bubble-there were about 20 million H100-equivalents of
compute in the world. Now there are 60 billion.
2026
20M
2030
100M
2032
1B
2034
60B
AI compute in the world at the beginning of each year in H100-equivalents.
mutually assured
compute destruction.
May 2029
Reliable Agent
Current Policy
In Discussion
Off The Table
62%
Employment
$50K
Median Income
2K
Researchers
0 mo
Total
Slodown
COMPUTE SELF-DESTRUCT SWITCHES
Datacenter have remote kill switches and backup destruction mechanisms.
Current Policy
US/China chip audit
Pause on AI R&D
Safety regulation for large
AI companies
Training rank
Baseline AI Company
Transparency
In Discussion
Total Research Transparency
Cap and Trade for robots and
AI self-destruct
AI
Off The Table
Universal Basic Galaxies
Citizen's Dividend
how we treat
nuclear weapons today.
Each superpower
data centers
on rival soil,
and leaves them
purposely undefended.
cheats
deal collapses,
the other
FOLLOW
destroy the chips
FOLLOW
before
to race ahead.
Plan A
Buy Time Slow down whenever is needed to have high confidence in safety.
Total Research Transparency Make almost all AI research fully visible to the public.
Diffuse AI Broadly Many companies in many countries at the frontier.
Reversibility Limit algorithmic progress; maintain Mutually Assured Compute Destruction.
also demands
total research transparency,
making almost
frontier AI research
public,
so any nation
would immediately
In this scenario,
The US and China
pause frontier AI training
in 2029,
scale slowly together
and delay
superintelligence
until 2040.
AI Futures Project
What is Plan A?
Why did we write this?
Why a scenario?
AI timelines?
the authors
stress this
recommendation,
not a prediction.
THE AI 2040 SCENARIO
AI companies are racing to build Als
that are smarter than humans in every way.
In AI 2027, we predicted that this would result in either extinction or irreversible
concentration of power. Plan A is our positive vision for what should happen instead.
In this scenario, humanity delays the development of superintelligence until 2040, makes all
AI research public, allows dozens of companies globally to catch up to the frontier, and
intentionally enacts a regime of mutually assured compute destruction.
Plan A is primarily a recommendation, not a prediction. This scenario is our best guess
as to what the future will actually look like. Instead, it's a vehicle for communicating and
stress-testing our policy recommendations.
In this AI 2040 scenario, Plan A is implemented successfully, albeit imperfectly and only in
the nick of time.
AI-driven extinction
somewhere between
10 and 30 percent.
The warning
Right now,
no one has
an adequate plan,
2029: Choose a Path
Race through the intelligence explosion by having AI self-improve and
putting them in charge of more things (datacenters, factories, weapons)
faster than China can?
"What? No, that's crazy!"
Slow down at least a bit for safety
and governance?
"Yes, and that's good
actually and/or so hard to do, there."
Let's make a deal with China.
But what?
(Two of many possible deals)
PLAN D
Race to AI
PLAN C
Burn the lead
PLAN B
Fight China
PLAN A
Verified Handover
PLAN E
Shut it all down
Sep 2029
PLAN C
Reliable Agent
Current Policy
In Discussion
Off The Table
62%
Employment
$50K
Median Income
2K
Researchers
0 mo
Total
Slodown
Jan 2030
Automated coder
clickng through the
choose-your-own-path
adventures.
the truth is
that nobody knows
how the next
plays out,
the projections
who've been on
the inside
some of the best
attempts at mapping it.
stay in the loop
with the latest
in AI and robotics...
AI and robotics...
argument
Follow
Rowan Cheung
@rowancheung
23
Following
53.5K
Followers
1M
Likes
Follow
Message