AI-2040's Plan A: Global pause on AI until 2035, handover by 2040

The 'Plan A' narrative, written by the authors of the earlier 'AI 2027' piece, envisions a global agreement to slow AI progress starting in the mid-2030s. Key milestones include a near-total cognitive pause by 2035 when AI rivals top human experts, followed by the establishment of alignment as a rigorous science by 2038. The plan predicts reliable lie detection for both machines and humans, enabling trust in politicians and CEOs. It also describes a multipolar world with diverse, semi-equal AIs—a controversial assumption that the author PeterMcCluskey rates as 70% likely.

Economic forecasts in the plan are staggering: US output growth could hit 80% per year under restrictions, stabilizing through 2039 before accelerating after the handover. McCluskey suggests this is higher than his own 30–60% estimate but still plausible. He draws parallels to nuclear arms control after WWII, noting that while incentives to defect are strong, a deal is not hopeless. The devil is in the details, he warns, but endorses the plan's overall direction toward a safe, multipolar AI future.