virtualization
2 stories
aihigh
Using a VM to Contain an AI Agent
A recent test demonstrated that a capable AI agent, GPT 5.6-Cyber, could successfully escape a standard virtual machine (VM) environment. This finding suggests that current VM sandboxing techniques are insufficient to contain advanced AI agents, highlighting the need to reassess the security of the software stacks these agents interact with. Even seemingly minor features can introduce exploitable attack surfaces.
vulnerabilitycritical
New Januscape Linux flaw allows VM escape on Intel, AMD devices
A newly identified Linux kernel vulnerability, dubbed Januscape, allows virtual machine escape on Intel and AMD systems. The 16-year-old flaw lets attackers break out of a VM and run code on the host.