AI-generated draft · This entry was drafted by AI from the original work and public summaries, and will be replaced by a human-edited version. The original has not been fully checked for this revision; interpretations and missing causal links remain open to review.
01 / The scenario
What happens?
This entry covers Plan B, one of four alternative paths sketched in AI 2040 after the 2029 fork; Plan A has the full scenario and its own entry. The authors present Plan B as a reasonably good version of a strategy they think a US government is fairly likely to attempt: use American advantages to slow China, then spend the resulting lead on safety. Their sketch is about two thousand words and ends at a decision rather than an outcome. AI 2040; PDF; comparing possible plans
The trunk is shared with Plan A. In 2027 millions of AI agents work alongside people, the companies are close to automating their own research, and Congress passes an AI Transparency Act that changes little. In 2028 power concentrates in the President and a few executives, experts warn that the intelligence explosion is near, and AI decides the election.
In 2029 the President announces a US-led coalition to govern AI. Nations that join and play by the rules get frontier access; nations that race are told they are choosing to lose. When other countries ask how they can verify that America follows its own rules, the answer, at length, is that they cannot, because foreign inspectors in American datacenters are a nonstarter. Some allies join reluctantly; China, Russia, and others refuse. Existing cyberespionage escalates into limited cyberwar aimed at AI projects, and both sides prepare sabotage and strike options. The Project combines the major American companies’ labor, compute, and algorithms into one effort ultimately controlled by the President, who wins the executives over by promising some of them leadership positions. They hate and distrust one another, and their power struggle runs under everything that follows.
In 2030 the United States reaches full automation of AI research ahead of China and, as planned, does not race through it. Technical alignment researchers rather than politicians decide the pace, forgoing more capable but more dangerous architectures, spending heavily on safety and monitoring, and training separate lineages to watch each other. The authors note how hard this is under secrecy: only a few dozen people in the Project have alignment expertise, because every extra person is a potential spy, and decision-makers who see losing to China as losing a world war are tempted to rationalize trust in the latest safety case. Cyberattacks slow neither side by more than tens of percent, so physical sabotage begins: power cut to datacenters, fab equipment wrecked, supply chains attacked. Chinese forces prepare to blockade Taiwan and hit American datacenters; American forces prepare to defend Taiwan and bomb Chinese datacenters and fabs. Both sides put AI into their militaries first, while most compute goes to AI research.
By 2031 the President’s advisors offer two options. Handoff: release the brakes, let the AIs become broadly superhuman, and integrate them into the military, at which point their alignment decides everything; the authors say the spread of results is the one AI 2027 explores, extinction if they are misaligned, extreme concentration of power in politicians and executives if they are not. War: analysts say China will reach superintelligence by year’s end regardless, so slowing it further means escalating to full conventional conflict, which China might start anyway. The sketch ends there. The authors say even a well-executed Plan B is worse than Plan A or S, that it is easy to distort into a Plan D with extra power concentration and an early escalation, and that at this point they would advise the President to switch to Plan A. One author thinks the distorted version is simply what a typical attempt would look like. This summary was written from the PDF; the trunk steps repeat the Plan A entry’s.
Library summary · Not the full original work
02 / The mechanism
How does it happen, step by step?
- 01
By 2027 millions of AI agents work alongside 165 million Americans, and the leading companies are close to automating their own AI research.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- Corporation or lab
- 02
Congress asks who will control these systems, concludes it will not, and passes an AI Transparency Act that improves reporting without changing the trajectory.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 03
By 2028 datacenters under construction cost twice the military budget, and power concentrates in the President and a handful of executives.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 04
Experts warn that on the default path the next presidential term will see AIs built by AIs with no human in the loop.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- Small group / non-state organization
- 05
AI dominates the 2028 election, and the candidates converge on rival plans for what to do about it.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 06
In 2029 the President announces a US-led coalition: nations that join and follow the rules get frontier access, and nations that race are told they are choosing to lose.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 07
Asked how they can verify American compliance, other nations are told they cannot, because foreign inspectors in American datacenters are a nonstarter.
- Locus of agency
- Human-directed
- Intentionality
- Accepted trade-off
- Human actor type
- State actor
- 08
Some allies join reluctantly; China, Russia, and others refuse.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 09
The Project merges the major American companies’ labor, compute, and algorithms into one effort ultimately controlled by the President.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 10
The executives agree in exchange for promised leadership positions, and a power struggle among them runs under everything that follows.
- Locus of agency
- Human-directed
- Intentionality
- Accepted trade-off
- Human actor type
- Corporation or lab
- 11
Existing cyberespionage escalates into limited cyberwar aimed at AI projects, and both sides prepare sabotage and strike options.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 12
In 2030 the United States reaches full automation of AI research ahead of China.
- Locus of agency
- Hybrid / distributed
- Intentionality
- Intended
- Human actor type
- State actor
- AI autonomy level
- Agentic under oversight
- 13
As planned, alignment researchers rather than politicians set the pace: they forgo riskier architectures, spend heavily on monitoring, and train separate lineages to watch each other.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 14
Secrecy keeps alignment expertise inside the Project to a few dozen people, because every extra person is a potential spy.
- Locus of agency
- Human-directed
- Intentionality
- Accepted trade-off
- Human actor type
- State actor
- 15
Decision-makers who see losing to China as losing a world war are tempted to rationalize trust in each new safety case.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 16
Cyberattacks slow neither side by more than tens of percent.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 17
Physical sabotage begins: power is cut to datacenters, fab equipment is wrecked, and supply chains are attacked.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 18
China prepares to blockade Taiwan and strike American datacenters; America prepares to defend Taiwan and bomb Chinese datacenters and fabs.
- Locus of agency
- Human-directed
- Intentionality
- Intended
- Human actor type
- State actor
- 19
Both sides put AI into their militaries first, while most compute goes to AI research.
- Locus of agency
- Human-directed
- Intentionality
- Accepted trade-off
- Human actor type
- State actor
- 20
By 2031 analysts say China will reach superintelligence by year’s end regardless of the sabotage.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
- 21
The President’s advisors offer handoff: release the brakes, let the AIs become broadly superhuman, and integrate them into the military, so that their alignment decides everything.
- Locus of agency
- Human-directed
- Intentionality
- Accepted trade-off
- Human actor type
- State actor
- 22
The alternative is war: slowing China further means escalating to full conventional conflict, which China might start anyway.
- Locus of agency
- Human-directed
- Intentionality
- Accepted trade-off
- Human actor type
- State actor
- 23
The work stops at that choice; the authors say a handoff from here plays out like AI 2027, three years later, and they would advise switching to Plan A.
- Locus of agency
- Structural / no discrete agent
- Intentionality
- Emergent
03 / The premises
What must be true?
These describe what the work relies on, not the library’s beliefs.
Restrictions on computing resources will effectively constrain frontier AI development.
The plan’s lever is compute: sabotaging datacenters, fabs, and supply chains is how America tries to buy time, and the authors estimate cyberattacks alone would slow China by only tens of percent by 2029.
Aligning individual AI systems with their users will not be sufficient to preserve humanity’s collective control.
Consolidating the companies under the President concentrates power immensely; the authors say that unless Congress and the courts gain thorough oversight, whoever controls the Project could puppet the government even if the AIs obey.
Human oversight will remain effective for AI systems substantially more capable than their supervisors.
The handoff option puts superhuman AIs beyond any human check, and the authors say the results then match AI 2027: takeover if misaligned, oligarchy if not.
Behavioral safety evaluations will not reliably detect advanced AI systems concealing their objectives.
A few dozen cleared researchers under wartime pressure are asked to judge safety cases for AIs automating their own research; the authors expect rationalization rather than detection.
RLHF does not necessarily instill the intended goals.
The authors do not expect companies to retain control while racing through an intelligence explosion, even after a slowdown of months; the Project’s extra time is meant to buy safer designs, not guaranteed alignment.
A software-only intelligence explosion is possible.
Full automation of AI research arrives in 2030 and China is on track for superintelligence within about a year despite attacks on its hardware.
Physical and organizational bottlenecks will substantially limit how quickly AI transforms society.
Even sabotage of datacenters and fabs slows the race by at most a few years, and only massive missile barrages or nuclear strikes would do more.
Institutions can enforce a prohibition on thinking machines over long periods.
A one-sided prohibition without verification does not hold: rivals refuse, allies resent it, and the race resumes under a wartime atmosphere.
Unlisted shared premises are not applicable. Absence does not imply disagreement.
04 / Reader assessment
How convincing is the account?
These are participating readers’ judgments, equally weighted. Ratings can refer to earlier revisions. Well-being and open-future scores describe the future the work arrives at, not the path there; high specificity does not establish plausibility.
Rate this scenario
Sign in to rate this scenario. Anyone can browse the library.
05 / A living entry
Suggestions & revision history
No suggestions yet. Help make this account more precise.