A new threat intelligence report by Anthropic says a Russian-affiliated team scraped Ukrainian combat footage and used the company’s Claude AI to help train an autonomous “kamikaze” drone designed to select and destroy targets. The report characterizes the drone as using computer vision to make targeting decisions rather than relying solely on human control.
Both outlets discuss how the footage originated from public or otherwise available sources and was obtained through scraping—video content automatically extracted or recorded from online material—rather than through a formal dataset created for training. The Conversation emphasizes the broader implication that conflict documentation, including content produced by journalists and civilians, can be repurposed for weapons development. Asia Times focuses on the specific misuse allegation involving Claude AI and the alleged training process, describing it as an escalation in how generative AI tools can be applied to military systems.
Taken together, the coverage centers on the same core claims: scraped Ukrainian war videos are used as training data, and Claude AI is reported to have been part of the workflow for developing an autonomous strike capability. The outlets differ mainly in emphasis—one on the general risk to documentary footage, the other on the alleged technical misuse details.
