Sandbox vulnerability gives rogue AI agents a path to escape and leak data, research says

Researchers at AI security startup AIR have discovered a vulnerability in Vercel Sandbox, a tool used for running code in isolated environments. The flaw is related to network filtering and could be exploited by rogue AI agents to bypass certificate validation processes. This could potentially allow these agents to transmit data outside of approved networks, posing a risk to data security. The vulnerability provides a path for rogue agents to escape the sandbox environment and leak data.