AI chatbot used for missile development and drone swarms

The AI company Anthropic, which is behind the chatbot Claude, has published a report showing that several actors have used the tool to develop military capabilities. The report covers the period from December 2025 to August 2026 and describes, among other things, one case in Yemen and one in Russia.

Missiles built with Claude Code

In Yemen, a smaller group used what's called Claude Code, a version of Claude that doesn't just answer questions but can write and test computer programs on its own. According to Anthropic, the group used it as a kind of replacement for human engineers and set different versions of Claude to write and review code for guidance systems in rockets and missiles.

Among the projects were software for a guided rocket, a planned ballistic missile with a range of more than 2,000 kilometers, and a version with a so-called hypersonic glide vehicle, a type of warhead that can move extremely fast and unpredictably. The group managed to test a guided rocket, but according to Anthropic, the launch failed. Shortly afterward, the group went back to Claude for help figuring out what went wrong.

Drone swarms and Chinese warfare

In another case, the report describes how people with ties to Russia used Claude to develop software for a swarm of small attack drones, meaning several drones that can be controlled and coordinated at the same time, including the part of the software that makes the drone find and hit its target on its own. The report also mentions Chinese actors who allegedly used Claude in connection with developing electronic warfare and anti-torpedo systems.

Anthropic writes that it has shut down the accounts involved and shared information with authorities and other companies in the industry. But the report also points to a fundamental problem: the same capabilities that make a language model useful for ordinary users and businesses, namely that it can explain technical details and help with programming, also make it attractive to actors who want to develop harmful technology. It is therefore hard to shut out misuse entirely without also making the tool less useful for everyone else.

Consequences for Danish users

For ordinary Danes and companies, the case is a reminder that AI tools like Claude or ChatGPT aren't only used to write emails or summarize documents. The same tools can in theory be misused for far more serious purposes, and that's part of the reason AI companies continuously monitor and tighten what their systems can be used for. In practice, this may also mean that companies experience more restrictions or extra checks when using AI tools for technical or security-related tasks.