AI powered agentic attacks are scary, but agents are particularly “suggestible” allowing us to both detect and derail them