NEWPractise supported commands in browser simulations, alongside the lesson.Try a browser practice exercise.Try a lab♥ Codelivly is 100% free, no paywall. If it helped you, chip in to keep it running.100% free. Help keep it so.Support us
Preview · 30 min
Master AI agent autonomy. Learn the critical security differences between assisted and autonomous agents, and defend against goal misgeneralization and infinite loops.
In the previous lesson we defined the AI agent and its agentic loop. But the industry uses the word "agent" loosely: a helpful coding assistant and a fully autonomous system that manages cloud infrastructure both get the name.
For security this is not a matter of words. The difference is the spectrum of agency: how much human oversight there is, and how far the agent may act without per-step approval.
An assisted agent limits the blast radius of a hallucination or a prompt injection to one human-reviewed step. An autonomous agent adds the risk of cascading failures, resource exhaustion (Denial of Wallet) and unauthorised privilege escalation at machine speed. Where an agent sits on the spectrum decides how strict your trust boundaries and guardrails must be.
Create a free account to unlock the rest of this task, complete tasks, and earn Rep.
Sign in