Scott Alexander argues against stopping AI work unilaterally
In an essay that set off a debate among researchers, Alexander says demanding that safety-minded labs quit the race alone destroys the backup plan if a global pause fails.
Source published · In the Daily Brief of 2 November 2026
Filed under AI safety, AI policy
Scott Alexander published “Against SURA” on 31 October. He separates two plans. Plan A is a global pause on AI development. Plan C, for a world without one, has safety-minded leading companies spend their lead on alignment research just before anyone reaches superintelligence. SURA is his name for the view that companies, employees and countries should stop unilaterally and raise awareness whatever others do.
He does not oppose a global pause, which he keeps as the first choice. His objection is that unilateral exit hands the lead to less careful companies and strips safety staff from those that remain, damaging Plan C for little gain to Plan A. He cites forecasters at the AI Futures Project whose survival estimates for such scenarios run from 10% to 63%.
Boaz Barak of OpenAI and Harvard and Nathan Calvin responded on X, asking for clearer scenarios for the period after superintelligence and for a sharper distinction between costly unilateral restraint and plain acceleration.