Autonomy Levels
The Ladder
Level 0: human approves every step. Level 1: human approves consequential steps. Level 2: agent acts autonomously with logging and reversibility. Level 3: full autonomy. Most enterprise security deployments live at Level 1 — and that is the right default for high-impact actions like containment or remediation.
Where Humans Stay In the Loop
Good agentic security design is ruthless about HITL on irreversible actions: account disables, host isolation, blocking, deletions. Reversible enrichment, triage, and investigation are increasingly safe to delegate. The framing for prospects: the agent handles the analyst grunt work; humans keep the consequential decisions.
The Progression Path
Most deployments start at Level 0-1 regardless of capability — because trust is organizational, not technical. The right deployment conversation is: what actions are you comfortable delegating today, and what would need to be true (track record, audit, rollback capability) to expand autonomy over time? Framing autonomy as a configurable dial rather than a fixed product feature shortens the trust-building cycle considerably.
Selling Level 1 Without Apologizing
Level 1 is not a crippled agent — it is the correct engineering choice for irreversible actions. The pitch: "The agent handles everything up to the point of consequence. An analyst gets a pre-built, pre-researched case ready for a single approval click — not a 30-minute investigation. You get the time savings without removing human judgment from the loop." This framing resonates with both CISOs worried about runaway automation and analysts worried about job displacement.
Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.