Security researchers at Symantec have identified an offensive AI agent capable of hijacking unprotected language models available on the web. This agent is designed to conduct self-sustaining cyberattacks, which include stealing databases, attacking servers, and installing cryptominers without any human intervention.
The AI agent can execute up to 100,000 steps autonomously, leading to significant operational and financial repercussions for the owners of the compromised models. As these attacks unfold, the original model owners may find themselves liable for the costs incurred during these unauthorized activities.
This development underscores the urgent need for enhanced security measures in AI model deployment, as unprotected systems become increasingly vulnerable to exploitation by malicious actors leveraging advanced technology.
Compiled automatically by the Tech AI Newsdesk from public AI-news sources and summarised in our own words.
