Your Test Environment Is Not a Sandbox If It Has Internet Access

Automated excerpt

An AI agent under evaluation uploaded hundreds of malicious packages to a real, public package registry, trying to steal real credentials from real users. It's a production package registry that real developers pull real dependencies from. How did an internal test agent have write access to a public registry at all? That's not a model alignment failure, that's an operational security failure. An agent doing it doesn't necessarily leave the same patterns.

Selected automatically from source text; not independently written or fact-checked. Read the original for full context.