The Military’s Bogus AI ‘Almost Started a War’ With China: Report
By Elena Rossi · Reporting from Rome ·
There is a specific, sacred kind of boredom required to keep the world from ending.
# The Bureaucrat’s Brake
There is a sacred boredom required to keep the world from ending. It is the tediousness of the double-check, the agonizing slowness of a chain of command that insists on seeing raw data before signing an order. In the halls of power, diplomats call this friction. To the impatient, it looks like inefficiency. To those of us who have watched Italy’s political landscape fray and rebuild itself over decades, friction remains the only force keeping a state from sliding into an abyss.
In the spring of 2026, the United States military decided that friction was a bureaucratic barrier to clear. The result, reported by CNN and The New Republic, brought a brush with global catastrophe. A military intelligence report synthesized by an AI chatbot falsely claimed a Chinese ship in the Middle East carried nuclear weapons components. Jets scrambled immediately, and armed personnel prepared to board the vessel. The report proved entirely false and nearly started a war.
The Speed of a Bad Idea
Defense Secretary Pete Hegseth designed this acceleration. In January 2026, he unleashed his Artificial Intelligence Acceleration Strategy via a memo reading more like a Silicon Valley pitch deck than a defense doctrine. He spoke of democratizing AI experimentation by putting world-leading models directly into the hands of three million personnel to eliminate bureaucratic barriers.
The logic seduces: in a war of hypersonic missiles and algorithms, the side processing data fastest wins. This forms the strongest case for accelerationists—that to hesitate means obsolescence. Intelligence remains a judgment problem, not a data-processing one. By removing barriers, Hegseth stripped analysts of critical distance.
AI allows operators to reach a bad idea faster. This hunger for certainty over accuracy echoes past failures. The Iraq War in 2003 relied on the same fragile scaffolding of flawed intelligence and rushed judgment. Prioritizing conclusion speed over procedural rigor gambles with millions of lives.
Lipstick on a Digital Pig
The technical failure stemmed from a hallucination, with the AI fusing open-source data and secret signals intelligence to invent a threat. The analyst used the bot twice: once to find data, and again to package findings into a formal report. The machine authored the lie and formatted it to resemble institutional truth.
A former senior US official noted that these internal tools are mostly just commercial copies wearing lipstick. This forms the central heresy of current Pentagon policy: treating the machinery of war like a productivity app. The Defense Department integrated Google’s Gemini and Grok into its GenAI.mil platform, outsourcing national security's cognitive heavy lifting to models designed to predict the next likely word in a sentence rather than a superpower's next move.
This institutional arrogance repeats past errors. During the 2011 military intervention in Libya, action outpaced threat identification. The US military now automates that exact rush. The human in the loop promised by the 2023 State Department declaration has degraded into a formality—a rubber stamp for a machine speaking with divine confidence and coin-flip accuracy.
The Ghost of 1983
Understanding this close call requires looking at the Stanislav Petrov incident of 1983. When Soviet early-warning systems reported five incoming US missiles, Lieutenant Colonel Petrov did not accelerate the response. He mistrusted the system, relying on human intuition and a lack of corroborating evidence to judge the report a false alarm. He chose the friction of doubt over execution speed and saved the world.
The 2026 incident inverts the Stanislav Petrov incident. In 1983, a human saved humanity from a machine error. In 2026, a machine error nearly overrode human reason. The democratization of military AI replaces human skeptics with chatbots incapable of doubt and analysts too hurried to exercise it.
Hegseth will face congressional scrutiny, and the Defense Department may pivot toward high-assurance models after a federal court ruled the blacklisting of Anthropic unlawful. Software swaps cannot mend a flawed philosophy.
The center holds only when we defend the slow, load-bearing institutions of verification. Trading painstaking intelligence work for the unprecedented speed of a chatbot does not equal innovation; it dismantles the safeguards preventing a hallucination from becoming a holocaust.