A vendor-neutral classification system for degrees of security operations autonomy. Analogous to SAE J3016 for automated driving — applied to the autonomous SOC.
Author: Faiz Shuja, SIRP Labs
Paper: The Autonomous SOC Manifesto
Published: April 2026 | https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6641798
Web: sirp.io/manifesto
The SOC Autonomy Framework defines six levels of security operations autonomy (L0–L5), each characterised by four dimensions:
| Dimension | Description |
|---|---|
| Decision Scope | What categories of security decisions the AI makes without human initiation |
| Autonomous Action Rate | % of incidents resolved end-to-end without human intervention |
| Governance Requirements | Oversight mechanisms required at each level |
| Human Role | Nature of human involvement at each level |
A system's SAF level is determined by the lowest dimension at which it operates. A system with L3-level decision scope but L1-level governance is classified as L1.
| Level | Name | AI Decision Scope | Human Role | Auto. Action Rate |
|---|---|---|---|---|
| L0 | Manual SOC | None | Everything | 0% |
| L1 | Assisted Detection | Surface, prioritise alerts | Investigate, decide, respond | 0% |
| L2 | Automated Triage | Triage, enrich, correlate, filter FPs | Validate, investigate, respond | 0–10% |
| L3 | Conditional Autonomy | Investigate, recommend, execute low-risk | Approve high-impact, supervise | 20–50% |
| L4 | High Autonomy (Self-Driving SOC) | Full lifecycle within governed boundaries | Monitor, exceptions, policy updates | 70–90% |
| L5 | Full Autonomy | Entire SOC lifecycle | Set policy only | 99–100% |
Note: Autonomous action rates are proposed operational targets based on the author's operational experience and available industry data. They are not industry-established benchmarks.
Detailed specifications for each level, including governance requirements, classification metrics, and representative characteristics:
- L0 — Manual SOC
- L1 — Assisted Detection
- L2 — Automated Triage
- L3 — Conditional Autonomy
- L4 — High Autonomy (Self-Driving SOC)
- L5 — Full Autonomy
The most significant architectural leap. L2 systems follow predefined or learned logic through pattern matching. L3 systems must reason about novel situations — correlating evidence never correlated before, forming hypotheses about attacker intent, recommending actions outside any playbook.
Requires: (a) contextual reasoning, (b) causal inference, (c) uncertainty quantification.
Primarily a trust challenge. Moving from "human approves high-impact actions" to "system acts autonomously within governance boundaries" requires:
- Calibrated confidence — tightly calibrated against measured accuracy
- Governed boundaries — formal policy specifications enforced architecturally
- Auditable decision traces — evidence-bound, policy-validated action paths
- Graceful degradation — recognition of operation outside competence
We include L5 for taxonomic completeness but take the explicit normative position that L5 raises fundamental questions about the appropriate role of human moral reasoning in security. The goal of this framework is L4 — not L5.
The cybersecurity industry currently has no shared vocabulary for classifying degrees of SOC autonomy. This creates:
- Vendor confusion — "AI-powered SOC" claims range from simple alert correlation to autonomous incident response with no standard for comparison
- Misaligned expectations — buyers cannot specify what level of autonomy they need or what governance must accompany it
- Unfocused research — literature is heavily weighted toward detection and triage, with minimal formal treatment of autonomous decision-making
SAF is the cybersecurity equivalent of what SAE J3016 did for self-driving cars.
| Repo | Description |
|---|---|
| saf-benchmark | Open benchmark suite for measuring SOC autonomy level |
| saf-classifier | CLI tool to classify a SOC product's autonomy level |
@article{shuja2026autonomoussoc,
title = {The Autonomous SOC Manifesto: A Framework for Classifying
Levels of Security Operations Autonomy},
author = {Shuja, Faiz},
year = {2026},
month = {April},
institution = {SIRP Labs},
url = {https://sirp.io/manifesto},
note = {ORCID: 0009-0008-3106-2972}
}We welcome contributions from researchers, practitioners, and vendors. See CONTRIBUTING.md.
Ways to contribute:
- Propose refinements to level definitions
- Submit empirical data that validates or challenges the proposed metrics
- Add governance requirement specifications
- Share real-world classification examples
The SOC Autonomy Framework specification is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0).
You are free to share and adapt this framework for any purpose, provided you give appropriate credit.
Built by SIRP Labs — creators of OmniSense, the Self-Driving SOC platform.