Rebels used Anthropic’s AI bot to develop guided weapons, report says
TL;DR
Militants in Yemen sought to use Anthropic's coding tool to produce guided rockets and missiles, the company found in a new threat report, noting it was able to block some but not all of the requests. According to the report, the activity ran from December 2025 to August 2026 across Claude Haiku, Sonnet and Opus, with the operators using the models in place of human software engineers for guidance and flight-control code.
Nauti's Take
The upside: Anthropic caught the abuse itself, banned the accounts and published the case in detail, which is real progress for industry transparency. The risk is just as concrete, since some requests got through and coding models clearly lower the barrier to building weapons software.
For companies and regulators, abuse detection and usage monitoring belong in any AI strategy alongside the features.
Sources