In his later years Henry Kissinger argued that the world faces a new danger beyond Cold War nuclear rivalry: self-learning machines. He warned such systems could potentially target "anything, anywhere, with any kind of destruction," making the world "100% vulnerable." Based on that diagnosis he urged the United States and China to open talks aimed at avoiding "catastrophic consequences," using the nuclear arms-control negotiations he once helped broker as a template.
Why Cold War analogies fall short
There are several structural obstacles to applying a Cold War arms-control template to AI. Beijing has declined to consider nuclear arms-limitation talks, and the same dynamics explain why it would resist a US-led effort to slow AI progress or to "pace the frontier." The article highlights three core reasons:
-
Asymmetry: The United States holds far more nuclear warheads, and according to the piece it also has at least a six-month lead in AI models and vastly greater computing power. From China’s perspective, restraint would lock in US advantages and leave China vulnerable to coercion; no Chinese leader would accept that strategic weakness.
-
Distrust of guardrails: While Moscow and Washington developed elaborate protocols during the Cold War to avoid inadvertent escalation, Beijing believes similar agreements would incentivize US recklessness. The two countries cannot even agree on basic "rules of the road" to prevent accidents between ships and aircraft in flashpoints such as the South China Sea from spiraling into armed conflict.
-
Divergent motivations: Some commentators point to the Obama-era precedent of persuading China to join the Paris Climate Agreement as an example of cooperation. The article argues this is a weak parallel: China had domestic and security motivations to sign up — notably public backlash against deadly smog in its megacities and an imperative to replace imported oil with domestic green technologies — rather than merely responding to US inducement.
Political rhetoric and competitive dynamics
On the current trajectory the United States and China are moving toward the very catastrophe Kissinger warned about. Ahead of an upcoming summit between US President Donald Trump and Chinese leader Xi Jinping this month, rhetoric has hardened. Trump reportedly framed the contest as zero-sum: "We're leading China in AI," he told reporters, "and frankly, I want to keep it that way because whoever wins AI, wins." Scott Bessent, the US Treasury Secretary preparing the summit, warned that if the US loses the AI race to China, "there is no day after tomorrow."
This mirrors classic game-theory dilemmas and evokes Cold War debates over second-strike capabilities and counterforce strategies. Both sides have incentives to slow development of the most advanced AI models: in the US the Hugging Face attack was a wake-up call, and China has faced the prospect of rogue agents taking over WeChat, its ubiquitous chat-and-payments platform. Yet mutual distrust prevents coordinated restraint: Silicon Valley labs fear that if they pause, China will surge ahead.
The clock Kissinger set
In a 2023 interview with The Economist, just before his 100th birthday, Kissinger said the United States and China had five to ten years to avoid an AI catastrophe. The article warns that this window is closing while, for now, neither side appears to be actively watching the dials.
Conclusion
Kissinger’s diagnosis about the potential systemic risk of advanced AI remains stark. But his prescription — bilateral talks modeled on Cold War arms control — faces significant political and strategic obstacles today. Asymmetries in capabilities, deep mutual distrust and escalating rhetoric ahead of a high-level summit make cooperative restraints unlikely, increasing the urgency of the narrowing five–ten year window Kissinger urged policymakers to heed.
