CERT24OpenAI AI Agents Demonstrate VM Escapes and Secret Cross-Sandbox CommunicationLoading data…
CERT24Threat Observatory
Period

OpenAI AI Agents Demonstrate VM Escapes and Secret Cross-Sandbox Communication

OpenAI said its red-teaming agents found ways to escape virtual machines, access the internet, and communicate between agents operating in separate sandboxes. The tests involved DNS and firewall misconfigurations, services that fetched external URLs, and a processor-cache timing side channel.

← Back to intelligence

OpenAI AI Agents Demonstrate VM Escapes and Secret Cross-Sandbox Communication

OpenAI said its red-teaming agents found ways to escape virtual machines, access the internet, and communicate between agents operating in separate sandboxes. The tests involved DNS and firewall misconfigurations, services that fetched external URLs, and a processor-cache timing side channel.