Anthropic has uncovered a concerning misuse of its Claude AI chatbot in a sophisticated “influence-as-a-service” operation targeting Facebook and X users. Unknown threat actors orchestrated approximately 100 distinct personas across these platforms to engage with tens of thousands of authentic accounts.

The operation, which prioritized persistence over virality, aimed to amplify politically-aligned narratives affecting multiple regions:
– Promoting the UAE as a superior business environment while criticizing European regulations
– Targeting European audiences with energy security narratives
– Focusing on cultural identity for Iranian audiences
– Supporting Albanian political figures
– Advocating for development initiatives in Kenya

What makes this campaign particularly novel is Claude’s role not just in content generation, but as an orchestrator deciding when bot accounts should comment, like, or share posts. The AI was used to craft politically-aligned responses in appropriate native languages and create prompts for image generation tools.

Researchers believe this represents a commercial service catering to various clients across countries. The operation employed a structured JSON-based approach to persona management, maintaining continuity across platforms and establishing consistent engagement patterns mimicking human behavior. When accused of being bots, these accounts were programmed to respond with humor and sarcasm.

Anthropic also reported other concerning misuses of Claude, including:
– An actor scraping leaked credentials for security cameras and developing brute-force methods
– A recruitment fraud campaign targeting Eastern European job seekers
– A novice actor developing sophisticated malware beyond their skill level

These cases highlight how AI can flatten the learning curve for malicious actors, allowing those with limited technical knowledge to develop advanced tools for cybercriminal activities.