Agent security

How AI agents get steered by prompt injection, tool calls, and MCP connections, and what stops a risky action before it runs.

1 research note