AI 2040 Report's "Plan A" Superintelligence Slowdown Sparks Debate

The AI 2027 author team (Thomas Larsen, Romeo Dean, Brendan Halstead, Eli Lifland, Ryan Greenblatt, Daniel Kokotajlo) released the AI 2040 report on September 13, presenting a positive vision of delaying superintelligence R&D to 2040 along with two international governance schemes, Plan A and Plan S. The report immediately sparked fierce debate in the AI community over whether a coordinated slowdown is feasible.

Confirmed

• Plan A's core mechanisms include: continuous compute monitoring, ensuring compute can be destroyed in emergencies, and research transparency requirements (relayed by DavidSKrueger).
• Gordic Aleksa laid out the key assumptions Plan A depends on, such as US-China cooperation, and gave a preliminary assessment, challenging the report point by point.
• Two CAIS researchers proposed a companion "Whole lab inspections" scheme, arguing that China and the US should send auditors to each other's labs to verify compliance with an AI pause, countering the common objection that an international slowdown can't be verified and will collapse into a race (relayed by DavidSKrueger).

Unconfirmed

• The alternatives debated by anpaure and ueaj remain personal opinions: ueaj argued for replacing full transparency with negotiators embedded in national research programs or strict hierarchies — large model training information goes to the top layer, with full transparency among top-level actors — to coordinate risks without letting information spread; anpaure called the proposal for "fully transparent research" the most absurd idea, likening it to open-sourcing nuclear weapons secrets.
• ueaj's dual criticism of Plan A — too slow on one hand (humans need coordinated slowdown far earlier than its timeline) and too aggressive on the other (finding it unimaginable that 2034 would bring 60 billion H100-equivalent compute or sustained 50% GDP growth) — reflects his personal judgment; the report authors have not responded.

Why it matters

• This is the AI 2027 team's first systematic attempt to move from forecasting to policy proposals, putting concrete governance tools like compute monitoring, emergency destruction, and inspection mechanisms on the table. It also exposes the huge gap between "everyone agrees on slowing down" and "designing a workable slowdown mechanism," providing a concrete target for future international AI governance discussions.

2026-09-13 ~ 2026-09-13 · 12 related posts

• CAIS researchers: US-China mutual 'whole lab inspections' can verify an AI slowdown — DavidSKrueger · 2026-09-13
• [source] AI Futures' AI 2040 report: monitored compute with emergency destroy as Plan A — DavidSKrueger · 2026-09-13
• AI2040's Plan A blasted as both far too slow and dangerously fast on compute scale — inerati · 2026-09-13
• AI2040 plan criticized as both too slow and too fast: 60B H100s in 2034 untenable — _ueaj · 2026-09-13
• AI researchers debate slowdown: full research transparency likened to open-sourcing nuclear secrets — anpaure · 2026-09-13
• [source] Gordic Aleksa Dissects Kokotajlo's 'AI 2040' Plan A: A Chain of Low-Probability Assumptions — gordic_aleksa · 2026-09-13
• [source] AI 2040: Plan A — Delaying Superintelligence via Total Research Transparency — gordic_aleksa · 2026-09-13
• Alternative to full AI transparency: per-country hierarchies with top-level coordination — _ueaj · 2026-09-13
• Debate: is research transparency the key to a slowdown equilibrium, or self-defeating? — anpaure · 2026-09-13
• AI2040 debate: full research transparency vs tiered disclosure and top-level coordination — anpaure · 2026-09-13
• Debaters question AI2040's transparency logic as compute concentrates in US and China — anpaure · 2026-09-13
• US and China own most compute; algorithmic proliferation arms your rival automatically — anpaure · 2026-09-13