According to SentinelLabs experts Alex Delamotte and Jim Walter, “AkiraBot’s usage of LLM-generated email message material demonstrates the emerging problems that AI poses in defending websites against phishing attacks.” The revolving cast of domains used to promote the Akira and ServiceWrap SEO offerings are the easiest ones to prevent because there is no longer a consistent approach to the email message contents like there were with earlier campaigns selling the services of these firms.
AkiraBot succeeded by assigning the following function to OpenAI’s talk API using the formula” You are a valuable associate that generates marketing information.” The LLM was instructed to change the variables with the blog name that was provided at runtime by a rapid. In consequence, each message’s body included a simple description of the service rendered by the recipient website along with the name of the recipient.
AkiraBot used an AI Talk quick Credit: SentinelLabs
The researchers wrote that the resulting information also includes a brief explanation of the intended website, giving the impression that the message was carefully chosen. The advantage of using an LLM to generate each message is that the text content is unique, and spam prevention becomes more challenging than using a consistent message template, which can be algorithmically filtered.
To assess the success and failure rates, SentinelLabs obtained log data AkiraBot left on a site. From September 2024 to January of this year, according to one folder, more than 80 000 websites had been safely delivered special messages. In contrast, messages that targeted almost 11, 000 realms failed. OpenAI thanked the scientists and reiterated that using its ai violated its terms of service.
Story changed to change the title.