SentinelLabs reported (via Ars Technica) that spammers used OpenAI's GPT-4o mini chatbot to generate unique spam messages that were sent to over 80,000 websites and successfully bypassed spam filters.
According to researchers, this activity lasted for about four months, starting in September 2024, and delivered unique spam messages through contact forms and chat widgets embedded in targeted websites.
The AkiraBot framework automated the creation of these emails, using Python scripts for domain rotation and GPT-4o mini to create personalized content. The uniqueness of the messages made them easier to deliver, as they did not look like typical spam emails. The text of each message mentioned the recipient’s website, creating the impression of a personalized approach.
OpenAI terminated the attackers' account after receiving the data from SentinelLabs and emphasized that such use of its chatbots was against its terms of service.