Explores the security implications of autonomous agents — prompt injection, sandbox design, and trust boundaries.