Why a Baruch Plan for AI?
In 1946, facing nuclear annihilation, the United States proposed something radical: place all nuclear weapons, research, and facilities worldwide under a single international authority.
The Baruch Plan (Wikipedia) called for an International Atomic Development Authority with genuine enforcement power—and crucially, no veto rights for any nation. It took less than a year after Hiroshima for this to become official US diplomatic policy.
The intellectual groundwork came from the Acheson-Lilienthal Report, largely written by Oppenheimer himself, which convinced President Truman that unilateral action couldn't counter proliferation risks.
The plan ultimately failed. The Soviets counter-proposed the Gromyko Plan, and despite support from Einstein and Russell, no middle ground was reached. What followed was an arms race, near-miss disasters, and a global surveillance apparatus that has—so far—barely averted catastrophe.
The AI Parallel
We have a second chance.
ChatGPT launched a winner-takes-all race among superpowers—the same dynamic that followed Hiroshima. Today, as in 1946, the US holds a temporary lead but cannot control global AI supply chains. And as in 1946, immense risks may compel even great powers to consider ambitious global governance.
The parallels are striking:
A reckless race forcing underinvestment in safety
Consensus building among scientists, agencies, and heads of state
A narrow window before capabilities spiral beyond control
Support Is Building
Experts' Calls for a Baruch Plan for AI include Yoshua Bengio (most-cited AI scientist), Ian Hogarth (UK AI Safety Institute), Allan Dafoe (Google DeepMind), Jack Clark (Anthropic), Jaan Tallinn (Future of Life Institute), and Nick Bostrom.
A University of Maryland survey found 77% of US voters support a comprehensive international AI treaty. Hundreds of experts signed the AITreaty.org open letter last spring.
AI Labs' Calls for Democratic Global Governance have grown over the past 18 months—from the CEOs of Anthropic, OpenAI, and Google DeepMind.