The Most Comprehensive AI Security Platform

The biggest security risk in a coding agent usually isn't the model - it's the harness around it. HiddenLayer researchers Kieran Evans, Kenneth Yeung, and Kasimir Schulz walk through why the layers wrapping an agent (Claude Code, Cursor, Codex, and the tools, skills, and orchestration around them) have become the real attack surface - and why, in many modern agents, you no longer need a prompt injection at all. If your instructions land in a path the model already trusts, it just follows them.