Plan A: Former OpenAI researchers propose international AI transparency treaty to manage superintelligence race

Researchers with direct OpenAI experience published Plan A this week, a policy scenario proposing how an international treaty could manage the race to superintelligence without existential catastrophe. The document, hosted at ai-2040.com, draws on conversations with frontier lab executives, lawmakers, and national security experts to outline a 2040 timeline where coordinated global governance replaces covert competition.

The core proposal mandates full transparency in advanced AI research and development. Under Plan A, countries would share research progress to enable mutual verification of safety protocols and prevent unilateral breakouts. The scenario assumes implementation occurs successfully—though imperfectly—at the last possible moment before superintelligence-level systems arrive.

The authors frame Plan A explicitly as a recommendation for policymakers, not a probabilistic forecast of what will actually happen. They contrast it with four alternative scenarios (Plans B, C, D, and S), each representing different U.S. policy responses to superintelligence risk. These alternatives serve as stress tests for the transparency-based approach.

What stands out

1. 01Transparency as the enforcement mechanism. If all frontier labs operate in the open, nations can verify compliance with safety measures and detect attempts to race ahead unilaterally.
2. 02Superintelligence expected within 14 years. The 2040 endpoint implies the authors expect superintelligence-level systems to emerge by then under current scaling trajectories if left unchecked.
3. 03OpenAI insider perspective. The scenario incorporates direct experience from inside a leading frontier lab, suggesting firsthand knowledge of how labs approach AGI timelines and safety decisions.
4. 04 Plans B, C, D, and S represent the main alternatives to international coordination, allowing readers to compare outcomes across different governance models.