In a startling illustration of how quickly artificial intelligence can affect high-stakes decisions, a U.S. analyst relied on a chatbot that blended open-source data with classified signals to claim a Chinese merchant ship in the Middle East was carrying components for a nuclear weapons program. The assessment, later identified as AI-generated, almost triggered a naval interception that a source described as “almost started a war.” The incident surfaced after a special-operations command analyst asked the AI to parse the vessel’s manifest, prompting officials to pause the operation once the report’s origin was traced back to the model.
AI-generated intel nearly triggers US naval action
The Department of Defense’s push toward an AI-first posture, formalized in a January memo by Secretary of Defense Pete Hegseth, encourages experimentation with leading U.S. AI firms. Partnerships announced in February with SpaceXAI’s Grok and in May with NVIDIA, Microsoft and Amazon illustrate the ambition, yet the near-miss underscores the danger of integrating generative models into sensitive intelligence workflows. While the specific chatbot—commercial or government-built—has not been disclosed, the episode demonstrates that even vetted AI tools can synthesize inaccurate conclusions, especially when they merge publicly available information with classified signals without human verification.
Trump administration meets Anthropic over AI safety
On Sept 15, senior officials from the Trump administration gathered virtually with Anthropic’s Washington lead, Tom Brown, to discuss the growing “AI safety” agenda. Pentagon chief technology officer Emil Michael and Commerce Secretary Howard Lutnick participated, reflecting the administration’s interest despite President Donald Trump’s public dismissal of AI risks as “a hoax” on Sept 14. The meeting followed a 3,800-word essay posted on Sept 12 by Anthropic CEO Dario Amodei, urging regulation and a slower pace for the most advanced models. Amodei’s plea was echoed by OpenAI’s Sam Altman and SpaceXAI’s Elon Musk, prompting a rare bipartisan chorus that also featured Senator Bernie Sanders and former adviser Steve Bannon warning that AI firms are endangering human interests.
Legal and commercial fallout with Anthropic
Anthropic’s relationship with the U.S. government has been turbulent. Earlier in the year, the Commerce Department challenged the security of Anthropic’s Mythos and Fable models, while the Pentagon rejected the company’s terms for military use, labeling the stance an infringement on authority and subsequently blacklisting Anthropic as a supply-chain threat. The firm is currently pursuing a lawsuit in Washington court over that blacklisting. Brown, who negotiated the lifting of export controls on Anthropic’s top models during the summer, later appeared alongside Lutnick at a G20-type summit in September, praising President Trump’s support for data-center expansion even as the Pentagon maintained its blacklist.
Broader implications for defense AI adoption
The dual episodes—a near-war caused by a faulty AI report and high-level talks over Anthropic’s safety stance—highlight the tension between rapid AI integration and the need for rigorous oversight. DoD officials, including Michael, have publicly refuted claims from former Anthropic researcher Jacob Coxon that AI could “gamble with our lives,” emphasizing that safeguards are in place. Yet the push to embed AI across military operations, backed by lucrative contracts with companies like SpaceXAI, NVIDIA and Amazon, means that missteps can have geopolitical consequences. As the midterms in November approach, both political and defense leaders are forced to balance competitive advantage against the risk of AI-induced escalation.
