Plans for the AI transition
A shorthand for strategic scenarios, which distinguishes them by how much coordination exists when very capable AI arrives. The letters come from the AI Futures Project’s comparison of possible plans. Scott Alexander grouped them into two families in Against SURA:
- Plan A: a global pause. Countries agree to stop or slow frontier development, possibly for decades, and use the time for alignment research, human intelligence augmentation, or making the pause durable. See Can AI progress be paused.
- Plan C: burning the lead. No global pause happens and the race continues. If the leading developers take safety seriously, they pause among themselves for as long as their lead over the nearest less careful competitor lasts, and spend that time on alignment using near-superhuman AI. Variants differ in who holds the lead: a whole country acting together (C+), one leading company (C), or no one, with safety work done only when convenient (D).
The scheme also includes Plan B, where the leading country slows its rivals while it works on alignment, and other cases. Writers group these differently, so check which letters an argument means.
Why the distinction matters
Actions can help one plan and hurt the other. Raising public alarm and building verification tools help Plan A. Putting safety-focused developers in the lead and preparing them to stop in time help Plan C. The 2026 Against SURA debate, over whether safety-minded labs and their staff should stop building even if others continue, is largely about this tradeoff; see The SURA debate turns on how doomed Plan C is.
Plan C works only if a lead of months is enough, which depends on How hard is alignment and Can AI automate alignment research. Critics note that the two-plan framing hides intermediate outcomes, such as messy slowdowns driven by national security concerns rather than a formal treaty.
Related: Pivotal act, Compute governance.