NEW YORK – Artificial intelligence is moving into world of warfare, raising questions about how powerful tools like Claude could be used in military applications. As AI technology becomes more capable, concerns are growing over its potential role in developing weapons and shaping the future of combat.
AI models were supposed to make life easier. Now, a new threat report suggests it may also be making warfare, weapons development and covert influence campaigns faster, and far more sophisticated. Anthropic, the company behind AI model Claude, revealed a series of cases in which people allegedly attempted to use its technology for everything from missile and drone development to surveillance and political influence operations.
The company says the cases shows some of the sophisticated attempts yet to misuse AI, offering a glimpse into how quickly powerful AI tools could become part of modern conflicts and covert campaigns.
One of the most striking cases involved a group in northern Yemen that was reportedly working on guided rockets and missile programs. According to Anthropic, the group attempted to use Claude Code, the company’s AI programming assistant, as a substitute for human software engineers working on guidance, navigation and control systems.
The report says the group even conducted an experimental guided-rocket launch. The test was reportedly unsuccessful, after which the group returned to Claude seeking help in understanding what had gone wrong. Anthropic says its safeguards blocked many requests, but the actors allegedly tried to get around those protections by hiding their intentions and splitting their work across different sessions.
The report takes an even more dramatic turn when it describes an alleged influence operation targeting narratives surrounding the Muslim Brotherhood, the war in Sudan and United Nations accountability mechanisms. Anthropic says an individual created an AI persona called “Dead Shot” and used Claude as part of what the company describes as a coordinated global campaign.
The operation involved network of roughly 300 fake or unauthenticated social media accounts, along with an NGO created to resemble a genuine Swiss-based organization. The campaign reportedly produced human-rights material and statements intended for presentation at the UN Human Rights Council.
The investigation linked the operation, with a “high degree of confidence,” to UAE government officials. However, the company also stressed that it could not confirm whether the prepared statements or dossiers actually reached their intended targets.
The report also describes alleged AI-assisted weapons development linked to China. In one case, a group reportedly used Claude to prepare technical documents for an anti-torpedo system, including lengthy proposals, presentations and procurement material. Anthropic says the group also worked on fire-control software and testing systems.
Then comes another chilling example: an alleged Russian freelance team working on automated FPV kamikaze drone swarms. Anthropic said the project, reportedly known as “DronDoc” or “Serafim,” used Claude Code to develop and test drone software. The company says the system was being designed for highly automated operations, with onboard technology potentially capable of identifying targets and making attack decisions without direct human intervention.
Investigators also found evidence that the group was experimenting with real hardware and training computer-vision systems using footage from the war in Ukraine.
The AI giant says all of the activities described in the report were investigated and disrupted, with relevant information shared with authorities and, where appropriate, other AI companies. The company says the incidents have also been used to strengthen its security systems. But the bigger story goes beyond any single account or operation.
The cases described by Anthropic raise a troubling question: as AI becomes better at coding, analysing information and solving complex technical problems, how difficult will it be to stop bad actors from turning those same abilities toward warfare and covert operations?
This report shows a darker possibility, that the same technology designed to accelerate human progress could also accelerate the machinery of conflict.
