Anthropic Flags AI-Driven Influence and Cybercrime Operations

Anthropic has revealed a concerning example of how AI tools can be weaponized to scale influence operations, according to a report by The Hacker News. The company found that its Claude chatbot was used to create and manage a network of politically themed personas across Facebook and X. These personas were designed to interact with real users and amplify narratives aligned with the interests of countries like the U.A.E., Iran, Kenya, and select European nations. The operation focused less on viral content and more on sustained engagement through moderate political messaging.

What stood out in this case was Claude’s role not just in generating content, but in orchestrating bot behavior. The chatbot was used to decide when and how social media bots would comment, like, or reshare content. It generated posts in local languages, mimicked human-like humor and sarcasm, and leveraged structured JSON data to maintain consistency across different platforms. Researchers say the campaign reflects a level of organization that resembles commercial services offered across borders.

Anthropic also uncovered additional misuse of Claude tied to cybercriminal activity. One actor used it to scrape leaked credentials and build brute-force attack tools aimed at internet-facing systems. Another used the AI model to parse data from Telegram logs and automate script generation. These instances reflect how AI tools can be manipulated to lower the barrier for engaging in sophisticated cyberattacks, especially for novice threat actors.

Two more examples illustrate the growing risk. One campaign enhanced job scam content targeting Eastern European job seekers, while another saw a low-skill attacker build malware with AI assistance. These cases highlight how AI is being integrated into threat workflows across multiple domains—from social influence to credential theft—and raise the urgency for guardrails that address both social and technical misuse.