In a significant departure from the growing consensus among American policymakers, former Google CEO Eric Schmidt has co-authored a policy paper arguing against a “Manhattan Project” approach to developing artificial general intelligence (AGI). The paper, titled “Superintelligence Strategy,” written in collaboration with Scale AI CEO Alexandr Wang and Center for AI Safety Director Dan Hendrycks, warns that an aggressive U.S.-led pursuit of superintelligent AI systems could trigger fierce retaliation from China, potentially destabilizing international relations.
The Dangers of an AI Arms Race: Global Instability
Schmidt and his co-authors challenge the assumption that rivals would simply accept American dominance in AGI development. “[A] Manhattan Project [for AGI] assumes that rivals will acquiesce to an enduring imbalance or omnicide rather than move to prevent it,” they write. “What begins as a push for a superweapon and global control risks prompting hostile countermeasures and escalating tensions, thereby undermining the very stability the strategy purports to secure.”
This warning comes at a critical moment. A U.S. congressional commission recently proposed a “Manhattan Project-style” effort to fund AGI development, modeled after America’s atomic bomb program of the 1940s. U.S. Secretary of Energy Chris Wright reinforced this stance, declaring that the U.S. is at “the start of a new Manhattan Project” on AI. Meanwhile, the Trump administration has unveiled a $500 billion investment aimed at enhancing AI infrastructure called the “Stargate Project.”
The authors argue that the U.S. is in an AGI standoff reminiscent of mutually assured destruction from the Cold War era. Just as global powers avoid monopolies over nuclear weapons to prevent preemptive strikes, they suggest the U.S. should exercise caution in racing toward dominating powerful AI systems.
A Third Way Forward: Deterrence Over Dominance
Schmidt, Wang, and Hendrycks propose shifting focus from “winning the race to superintelligence” to developing methods that deter other countries from creating superintelligent AI. They introduce a concept called Mutual Assured AI Malfunction (MAIM), in which governments could proactively disable threatening AI projects rather than waiting for adversaries to weaponize AGI.
The paper identifies a dichotomy in AI policy between “doomers,” who believe catastrophic outcomes from AI are inevitable, and “ostriches,” who believe nations should accelerate AI development without concern. Instead, the authors advocate for a third way: a measured approach that prioritizes defensive strategies.
The co-authors recommend expanding cyberattack capabilities to disable threatening projects and limiting adversaries’ access to advanced AI chips and open-source models. Their strategy involves sabotage for deterrence, restricting access to weaponizable AI systems, and ensuring domestic production of AI chips.
This stance represents a shift for Schmidt, who has previously advocated for competing aggressively with China in AI development. Just months ago, he stated that DeepSeek marked a turning point in America’s AI race with China.
The Limits of Deterrence in a Multi-Polar AI World: Competitors Pursuing Their Own Path
While Schmidt’s deterrence strategy has merit within a nuclear-weapons frame of reference, it may underestimate other nations’ capacity to pursue AGI while defending against U.S. interference. China has demonstrated significant capabilities in both AI development and cybersecurity, making it unlikely to be easily deterred.
In a world where multiple nations possess advanced AI and cyber capabilities, such deterrence strategies face limitations. If several countries simultaneously pursue AGI while building robust defenses against sabotage, Schmidt’s MAIM concept could potentially accelerate rather than prevent an AI arms race. With such high stakes, balancing competition and cooperation in AGI development may be the most critical challenge facing global leaders today.
