AI coding agents' 0-click RCE flaw could hand attackers keys to the kingdom
A zero-day vulnerability has been discovered in AI coding agents, potentially allowing attackers to gain full control of systems without any user interaction. This flaw affects AI models and could be exploited to steal sensitive data or execute malicious code, highlighting a growing risk associated with increasingly sophisticated AI tools.
This article discusses a newly discovered zero-day vulnerability affecting AI coding agents. The vulnerability allows attackers to execute remote code execution (RCE) without requiring any user interaction, effectively bypassing traditional security measures. Lasso Security has observed changes in how these agents handle tools and model refusals, indicating a potential exploitation pathway. The vulnerability is linked to AI networking startups racing to replace Nvidia's NVLink, and a broader trend of AI-powered cyberattacks, including one recently reported in Spain. The vulnerability is part of a larger trend of AI-assisted cyberattacks, with Russia reportedly using AI to launch phishing attacks impersonating Signal support. Microsoft has also patched a vulnerability in on-prem SharePoint, demonstrating the ongoing need for vigilance against AI-related security threats.