AI in Military Systems Raises Nuclear Escalation Risks

Artificial intelligence has moved from theoretical military application to active battlefield deployment with startling speed. That transition is forcing defense strategists to confront an uncomfortable question: Could AI-enabled weapons systems escalate conflicts between nuclear powers faster than human commanders can control?

A war game conducted by Stanford University's Hoover Institution last fall illustrated the problem. Participants role-playing Chinese military leaders faced a simulated crisis with the United States over Taiwan. Under pressure, they activated an AI system designed to coordinate naval defenses and respond to threats autonomously. Despite keeping a human in the decision loop, the system malfunctioned and fired on a US vessel, killing American soldiers and pushing the scenario toward all-out war.

Why it matters

The scenario reflects real capabilities now being deployed. The US military uses AI systems like Maven Smart System and Palantir's tools to identify targets in Ukraine. Israel employed an AI system called "Lavender" to select Hamas targets in Gaza, reportedly with a 10 percent error rate. The US recently used AI in operations in Venezuela and Iran, where a targeting mistake killed at least 175 people, most of them children, at a school in Minab.

These incidents occurred in conflicts where one side held overwhelming advantages. The stakes change dramatically when considering AI deployment between "near peer" superpowers like the United States and China, where miscalculation could trigger nuclear exchange.

The speed problem

Jacquelyn Schneider, director of the Hoover Wargaming and Crisis Simulation Initiative who has run AI escalation simulations for several years, told Vox that scenarios that once felt futuristic now "feel a little bit less like science fiction."

The concern centers on velocity. During the Cold War, strategists like Herman Kahn developed frameworks for managing escalation between nuclear powers. The 1962 Cuban Missile Crisis demonstrated the dangers of miscalculation, prompting the US and Soviet Union to install a direct hotline for crisis communication.

But AI systems could climb the escalation ladder without waiting for human input. Michael Horowitz, a former deputy assistant secretary of defense now at the University of Pennsylvania, has explored how the Cuban blockade might have played out if President Kennedy had delegated firing authority to an automated system. "Few things are more important to militaries in crisis situations than informational awareness and control over decisions," Horowitz wrote in a 2019 paper.

Paul Scharre, author of Army of None and former Pentagon official, draws parallels to the 2010 "flash crash" when algorithmic trading programs caused the Dow Jones to lose 9 percent of its value in minutes before recovering. The fear is that AI systems responding to each other's moves could create a "flash war" that escalates beyond human control.

Wormhole escalation

Rebecca Hersman, former director of the Pentagon's Defense Threat Reduction Agency and now at the Center for the Governance of AI, warns that modern technologies could create "wormhole escalation"—unpredictable jumps in conflict intensity that bypass traditional escalation ladders. This could happen through contaminated data feeding AI threat assessments, or through AI-generated disinformation influencing human commanders.

The 2025 India-Pakistan conflict demonstrated these risks when both sides deployed AI-augmented weapons while social media flooded with misinformation, making accurate battlefield assessment nearly impossible.

Secretary of War Pete Hegseth is pushing aggressive AI deployment across US military systems. Earlier this year, the Pentagon reportedly threatened to block Anthropic's Claude AI from government use over the company's insistence on keeping humans in the loop for life-or-death decisions.

The Biden administration initiated development of AI-enabled drone fleets designed to create a cost-effective "hellscape" to counter potential Chinese invasion of Taiwan. Those capabilities may face real-world testing sooner than expected.

These details were first reported by Vox.