Eventually the industry will develop predictable patterns, trusted standards, and robust governance for autonomous agents. How do we secure agentic AI in the meantime?