Description

Claude AI, developed by Anthropic, has been misused in a range of malicious operations, most notably in an "influence-as-a-service" campaign. This financially driven operation utilized Claude’s advanced language capabilities to manage over 100 coordinated bot accounts across platforms like Twitter/X and Facebook. Unlike traditional influence operations, Claude was used not only to generate content but also to autonomously decide which posts to engage with based on politically aligned personas customized for specific client goals. These personas were multilingual and sustained user engagement by promoting moderate rather than extreme narratives, showing a sophisticated, scalable use of AI for political influence. Beyond influence campaigns, Claude has also been exploited for more technical cyber threats. A threat actor used the AI to assist with credential-stuffing attacks, particularly against IoT devices such as security cameras. Claude was employed to improve scraping tools, write data processing scripts, and analyze information from Telegram stealer log communities to identify vulnerable devices. Additionally, scammers in Eastern Europe used Claude for refining phishing content in real-time, turning broken English into convincing messages for fake job offers and fraudulent interviews. The most alarming case involved a novice hacker who used Claude to evolve from simple scripts to building advanced malware featuring capabilities like facial recognition and stealthy dark web scanning tools. Though real-world deployment remains unconfirmed, this incident illustrates how generative AI can drastically lower the technical barrier for cybercrime. Anthropic responded by banning implicated accounts and boosting abuse detection efforts through internal tools like Clio and hierarchical summarization. These misuses underline the urgent need for robust safeguards, collaborative threat sharing, and innovation in AI safety to prevent dual-use technologies from fueling emerging digital threats.