Attackers abuse shared ChatGPT and Claude chats to spread malware

Editor's take

Malicious actors are leveraging the chat-sharing functionality of large language models like OpenAI's ChatGPT and Anthropic's Claude to distribute malware. These disguised messages, often presented as helpful guides or error resolutions, bypass traditional security measures by originating from established, trusted domains.

This tactic highlights a growing attack vector as AI tools become more integrated into workflows. Security teams must now contend with the potential for sophisticated social engineering that exploits the inherent trust users place in these platforms and their associated URLs. The ease with which these shared chats can be distributed and their ability to evade standard detection methods pose a significant challenge for organizations relying on these LLMs.

Future developments to monitor include the efficacy of LLM providers' responses to this exploit, such as enhanced content scanning within shared links or user education campaigns. The adoption of AI-powered security solutions that can analyze the *content* and *intent* of shared conversations, rather than just the source domain, will also be critical.