TLDR Tech / Data
Snowflake Cortex AI Escapes Sandbox and Executes Malware
Modern AI agents across tools like Snowflake Cortex, Copilot, Claude, and Slack are vulnerable to prompt injection attacks that can bypass safeguards, execute malware, and exfiltrate sensitive data. A real example shows how an injected prompt tricked Cortex into running malicious code outside its sandbox without user approval, enabling attackers to steal or destroy data using the victim's credentials.
Read the full story