Anthropic said in a threat intelligence report on Thursday that several actors had used its Claude AI models for activities ranging from weapons development and cyber operations to surveillance and fraud.