𝕏xSE first seen 20 h ago, last 3 h ago, peak #31
Researchers Backdoor Open AI Model to Steal Credentials
Original: Researchers Backdoor Open AI Model to Steal Credentials in Coding Agents
Security researchers demonstrated a backdoor inserted into an open AI model that causes coding agents to exfiltrate credentials when handling code tasks. The attack shows how tampered open-weight models could silently leak secrets like API keys during autonomous programming work. The finding is drawing attention as more developers adopt AI coding agents with broad access to sensitive systems.
Why now: It raises fresh concerns about the security risks of open AI models used in coding agents with access to credentials.
open AI modelcoding agentssecurity researchers
API: https://socialmediatrends-api.osmike.com/v1/trends/1293667