Calls for slower, transparent development toward superintelligence, with temporary training pauses when verification or control is inadequate and cautious resumption afterward. [1][2]
Initial statement. First reviewed observation of the July 2026 Plan A recommendation and Kokotajlo’s later clarification of its overall pace.
Source [1] · Recorded interview transcript
AI 2027’s author returns with a plan to change the ending
Attributed to: Daniel Kokotajlo · 80,000 Hours
Published
“We are advocating for going slower than you could go at maximum speed.”
Source context. Recorded July 27–28 and published August 27. In his own words, Kokotajlo clarifies that the plan is slower than maximum speed yet continues AI development, deployment and diffusion, with temporary literal pauses. He says some elements of the scenario are illustrative predictions rather than fixed recommendations.
Scope
AI research and development toward superintelligence, especially recursive self-improvement
Conditions
Temporary pauses are used when verification or control is inadequate; otherwise development, deployment and diffusion continue cautiously.
Stated rationale
He argues that maximum speed would leave too little time to establish alignment and safety.
Source details
primary · Original 80,000 Hours interview transcript with recording and publication dates.
Discovered 28 Sep 2026.
Accessed 28 Sep 2026.
Source [2] · Coauthored policy proposal
Summary of Plan A
Attributed to: Daniel Kokotajlo · AI 2040 / AI Futures Project
Published
“Temporarily pause AI training while developing better verification technology.”
Source context. The official summary names Kokotajlo as coauthor. The authors’ dated July 9 announcement calls the plan a recommendation, not a prediction. It proposes buying time, a temporary training pause to build verification, and resumption under safety-case regulation and research transparency. Its fictional future timeline should not be treated as a fixed deadline.
Scope
Frontier training and eventual superintelligence development
Conditions
Resume R&D under safety cases, research transparency and improved verification; slow when safety confidence requires it.
Stated rationale
The coauthors seek time to verify safety and avoid a dangerous race.
Source details
primary · Original official Plan A summary; release date cross-checked with coauthor announcement dated July 9.
Discovered 28 Sep 2026.
Accessed 28 Sep 2026.
Editorial assessment. His own interview explicitly defines the overall plan as slower continued development. Plan A’s temporary pauses are essential steps, not a permanent stop or an unconditional present moratorium. A separate hypothetical Plan S is a contingent fallback, not the headline recommendation. Do not convert the scenario’s calendar into a fixed real-world deadline.