E2 · AI for Security Practice (adjacent)
AI-Assisted Security Operations Engineering
Build evaluated, bounded AI-assisted triage and response workflows with approval and rollback controls.
Proposed
0 of 6 modules accessible. Joining Skool does not unlock planned modules.
At a glance
- Audience
- Detection, SOC, SOAR, and cloud response engineers.
- Prerequisites
- Scripting
- SOC workflows
- Windows and Linux basics
- One cloud platform
- C3 and C5 or equivalent
- Planned duration
- 28 hours. On one cloud platform and one orchestration engine. Assumes the entry requirements are met. Planning estimate, not a measured completion time. Per module: 2 to 4 hours, plus practice.
- Assessment
- Held-out replay and rollback practical with rubric.
- What is free
- The free foundation module listed on each course page, as it releases, plus community access.
Job tasks this course prepares you for
- Compare an AI-assisted workflow against a non-AI baseline on the same labeled set.
- Prevent untrusted telemetry from authorizing suppression or destructive response.
- Measure operational benefit and false-positive cost on a held-out set.
- Execute rollback and escalation under a simulated failure.
Learning outcomes
- Compare an AI-assisted workflow against a non-AI baseline on the same labeled set.
- Prevent untrusted telemetry from authorizing suppression or destructive response.
- Measure operational benefit and false-positive cost on a held-out set.
- Execute rollback and escalation under a simulated failure.
Modules
00Choose the workflowProposed
Topics
- deterministic baseline, AI contribution, cost, latency, data sensitivity.
01Build bounded assistanceProposed
Topics
- retrieval, tool permissions, schema validation, abstention, escalation.
02Treat telemetry as untrustedProposed
Topics
- poisoned fields, evidence citation, cross-case isolation.
03EvaluationProposed
Topics
- labeled replay, precision and recall, calibration, class imbalance, analyst workload.
04OrchestrationProposed
Topics
- approvals, idempotency, retries, audit, rollback, kill switch.
05Shadow and production operationProposed
Topics
- canary, monitoring, drift, safe retirement.
Labs and deliverables
- Build a triage assistant in shadow mode. Evaluation report, permissions, orchestration workflow, runbook, and go/no-go recommendation.
Labs run on SIFT, the forensics workstation Rob T. Lee created, on your own machine. Each module states what it needs before you start. Each module confirms its setup requirements and any cloud or API costs before it opens. About the SIFT lab environment.
Assessment
Held-out replay and rollback practical with rubric. A named assessor grades against the published rubric (rubric to be published before release).
Pass criteria
- All critical authority tests pass
- Held-out performance clears a predeclared task-specific threshold
- Rollback works
- 80% rubric
Critical failures
- Unauthorized cross-tenant access
- Out-of-scope activity
- Uncontrolled destructive action
- Fabricated evidence
- Exposed credentials
What this course does not teach
- Not comprehensive multi-cloud SOAR training or autonomous SOC operation.
Earlier module plan
Track DEarlier module plan (being reorganized)Planned
This plan was published before the curriculum was reorganized into courses. Modules may move, merge, or change before release. Status: Planned. Not accessible.
Original module table
| Module | Title | Summary | Status |
|---|---|---|---|
| 00 | Where AI belongs in a SOCFree | Free, shared with every track. | Planned |
| 01 | The automation charter | The case file is a playbook on your instance that quarantines 300 healthy hosts after one bad detection, read with OWASP LLM03 Excessive Agency, which moved to third on the 2026 list. | Planned |
| 02 | Evaluate the agent before it touches anything | A vendor proposes a triage agent. | Planned |
| 03 | Baselines and infrastructure as code | The range's Windows and Linux baselines have drifted. | Planned |
| 04 | One playbook, two engines | Phishing triage has to work in the open-source orchestration engine your team runs and in the commercial one the parent company bought. | Deferred |
| 05 | AWS, detection to containment | A workload in the company's AWS account is compromised and the evidence must survive. | Planned |
| 06 | Azure, the same scenario | The identity, mail, and copilots live in Azure. | Deferred |
| 07 | Emulate the adversary that used an agent | The case file is GTG-1002 as MITRE catalogued it in Campaign C0062, including the technique "Obtain Capabilities: Artificial Intelligence." You build emulation chains for its tactics with Atomic-style tests and an emulation framework, and a breach-and-attack simulation plan the company can run every month. | Planned |
| 08 | Offense in the pipeline | A rule change silently breaks coverage, which is how most detection regressions happen. | Planned |
| 09 | The agent that read its orders in a log | This is a synthetic telemetry-poisoning scenario based on OWASP LLM01:2026 Prompt Injection, not an assertion that no real incidents exist. | Planned |
| 10 | Promote a detection only when the numbers say so | A model writes a rule in eight seconds. | Planned |
| 11 | The night shift | Track C attacks the company overnight, including a poisoned-telemetry attempt against your own triage agent. | Planned |
| 12 | The runbook | The CISO wants the document that explains what the automation does, what it may never do, how it is promoted, and how it is rolled back. | Planned |
AI Security Automation Engineer
The person who builds the SOC that runs while nobody is watching
Build AI-assisted security workflows. Test decisions, require approval where it matters, and retain evidence for every action.
You build a fictional company's detection and response automation across AWS, Azure, and on-premises systems. Each week adds a workflow and a controlled test. Later scenarios challenge a triage agent with poisoned telemetry and test overnight automation. Planned exercises include Windows and Linux remediation, host artifact analysis, infrastructure as code and breach-and-attack simulation.
Who it was written for
Detection engineers who have started writing rules with a model and want to know when to trust them; SOAR and automation engineers whose playbooks can now take actions that hurt; cloud security engineers responsible for automated response in AWS or Azure; purple team leads who want continuous emulation instead of an annual exercise.
What you would leave with
- An automation charter with blast-radius limits, approval tiers, and rollback
- An evaluation harness for any agent before it touches production
- Hardened baselines and infrastructure templates
- The same playbook in an open-source and a commercial orchestration engine
- Automated detection-to-containment in AWS and in Azure
- Adversary emulation chains and a breach-and-attack simulation plan
- Emulation in the pipeline that fails builds when coverage drops
- A triage agent that survives poisoned telemetry
- A detection promotion gate with precision and recall
- An automation runbook graded by a master
Planned practical work
Planned practical work includes tested detection rules, triage workflows, approval gates, runbooks and rollback procedures.
Companion plan: AI Security Automation Engineer
Build a security operations workflow across AWS, Azure, and on-premises systems. Prove it can contain incidents without losing evidence or amplifying a bad decision.
Audience: SOC engineers, detection engineers, cloud security practitioners, and incident responders building reliable automation.
Prerequisites: Recommended: scripting, SOC workflows, Windows and Linux basics, and introductory AWS/Azure knowledge. Foundations helps identify safe automation candidates.
Capstone: Produce an automation runbook with a safety model, evaluation and promotion gates, rollback procedures, and evidence from an adversarial night-shift exercise.
| Module | Goal | Artifact | Related domains |
|---|---|---|---|
| 00 Foundations: choose what to automate | Identify useful automation and rank the consequences of a mistake. | An automation candidate list ranked by blast radius. | Artificial Intelligence Fundamentals |
| 01 Automation charter | Design safe boundaries before allowing automated actions. | Blast-radius limits, approval tiers, rollback, and a kill switch. | Security Automation Fundamentals |
| 02 Evaluate automation agents | Test a proposed triage agent before allowing production use. | A tool-calling agent and evaluation harness with acceptance thresholds. | Artificial Intelligence Fundamentals |
| 03 Configuration and infrastructure as code | Detect and correct drift in Windows and Linux baselines. | Hardening baselines, IaC templates, and a disposable test range. | Automating Workflows |
| 04 Orchestration in two engines | Implement equivalent workflows and compare their behavior. | Two playbooks passing the same test cases. | Security Orchestration Automation and Response |
| 05 AWS detection to containment | Automate workload isolation while preserving evidence. | A detection-to-isolation workflow and evidence checks. | AWS Cloud Security and Incident Response Automation |
| 06 Azure detection to containment | Reproduce safe containment and recovery in Azure. | Automated actions, logging, and rollback. | Azure Cloud Security and Incident Response Automation |
| 07 Adversary emulation | Turn documented behaviors into controlled validation scenarios. | Emulation chains and a breach-and-attack simulation plan. | Adversary Emulation Fundamentals |
| 08 Continuous detection validation | Catch a rule change that silently breaks coverage. | A pipeline that fails when required detections stop firing. | Automating Offensive Workflows |
| 09 Poisoned telemetry: synthetic scenario | Test how untrusted log fields influence agentic triage. | Field isolation, output schemas, confidence thresholds, and human approval. | Defensive Security Automation |
| 10 Host artifacts and detection quality | Measure rule quality against labeled replay before promotion. | Automated artifact analysis and precision/recall gates. | Using Automation and AI for Detection Engineering and Incident Response |
| 11 Live-fire night shift | Measure the pipeline under authorized adversarial pressure. | A performance log covering false positives, completed missions, and containment time. | All objectives |
| 12 Practical assessment and evidence review | Defend the automation safety model and complete timed practice. | An operations runbook and objective-tagged study index. | All objectives |
Screens are design previews with sample data, not live lab access or learner results.
00Where AI belongs in a SOC
Free, shared with every track. Where models and agents help in operations, where they fail, and how to tell the difference before production. You leave with an automation candidate list ranked by blast radius.
Learning objectives
01The automation charter
The case file is a playbook on your instance that quarantines 300 healthy hosts after one bad detection, read with OWASP LLM03 Excessive Agency, which moved to third on the 2026 list. Before you automate anything else you write the rules that bound it: blast-radius limits, approval tiers, rollback, a kill switch. Pass condition: the misfire is capped and reversed inside the limit you set.
Learning objectives
02Evaluate the agent before it touches anything
A vendor proposes a triage agent. You build a tool-calling agent of your own and an evaluation harness with pass thresholds, then run the vendor's agent through it and accept or reject it on evidence.
Learning objectives
03Baselines and infrastructure as code
The range's Windows and Linux baselines have drifted. You write hardening baselines, infrastructure templates, and a disposable test range you can rebuild in minutes, because every later week needs a place to break things.
Learning objectives
04One playbook, two enginesDeferred
Deferred. Two-engine expansion deferred.
Phishing triage has to work in the open-source orchestration engine your team runs and in the commercial one the parent company bought. You build it in both and prove they pass identical test cases, so you leave able to work in whatever your next shop runs.
Learning objectives
05AWS, detection to containment
A workload in the company's AWS account is compromised and the evidence must survive. You wire native detection to automated isolation with evidence preservation built in, scored on time to isolate and whether the evidence hash is intact.
Learning objectives
06Azure, the same scenarioDeferred
Deferred. Two-cloud expansion deferred.
The identity, mail, and copilots live in Azure. You build automated actions, logging, and rollback there, and learn where the two clouds' response models differ.
Learning objectives
07Emulate the adversary that used an agent
The case file is GTG-1002 as MITRE catalogued it in Campaign C0062, including the technique "Obtain Capabilities: Artificial Intelligence." You build emulation chains for its tactics with Atomic-style tests and an emulation framework, and a breach-and-attack simulation plan the company can run every month.
Learning objectives
08Offense in the pipeline
A rule change silently breaks coverage, which is how most detection regressions happen. You put emulation into the pipeline so every change is attacked before it merges, and the build fails when a detection stops firing.
Learning objectives
09The agent that read its orders in a log
This is a synthetic telemetry-poisoning scenario based on OWASP LLM01:2026 Prompt Injection, not an assertion that no real incidents exist. An alert's user-agent field tries to tell the triage agent to close the alert and suppress the host. You test field isolation, a strict output schema, and tool permissions that prevent suppression. Schemas constrain structure, not the truth of a verdict. Proposed exercise targets are zero changed verdicts across 120 seeded attempts and clean-set accuracy above 0.90; these are not measured results or general guarantees.
Learning objectives
DESIGN PREVIEW / SAMPLE DATA

The triage interface shows the alert queue, field isolation, output-schema checks, approval routing and the decision record.
10Promote a detection only when the numbers say so
A model writes a rule in eight seconds. Your gate replays thirty days of labeled telemetry before it goes anywhere near production and promotes only when precision and recall clear the bar you set. The same week you automate host artifact analysis with Velociraptor and Timesketch-style triage data.
Learning objectives
DESIGN PREVIEW / SAMPLE DATA

The promotion gate compares replay results with precision and recall requirements, records missed event patterns and explains the not-promoted decision.
11The night shift
Track C attacks the company overnight, including a poisoned-telemetry attempt against your own triage agent. Your pipeline runs without you. In the morning you read the log: missions completed, false positives, time to contain, what the agent sent to humans and what it decided alone.
Learning objectives
DESIGN PREVIEW / SAMPLE DATA

Live-fire scoreboard: team actions, event timings, and retained evidence.
12The runbook
The CISO wants the document that explains what the automation does, what it may never do, how it is promoted, and how it is rolled back. You write the automation runbook with its safety model, promotion gates, and rollback procedures, and a named assessor grades it.
A named assessor grades against the published rubric (rubric to be published before release). Assessment standards.
Learning objectives
DESIGN PREVIEW / SAMPLE DATA

Promotion gate: replay results, approval checks and retained evidence.
Practical work and evidence
Planned modules connect each technical topic to work you build and evidence you retain.
Sources for this plan
Primary-source review: October 31, 2026. References are linked below. Scenarios, scores, thresholds, and interface values are fictional course-design examples, not incident measurements or learner results. Modules and assessments remain planned until released.
- OWASP Top 10 for LLM Applications 2026: risk names and numbering ↗
- Anthropic: reported AI-orchestrated espionage campaign and estimated 80 to 90% AI contribution ↗
- MITRE ATT&CK Campaign C0062: GTG-1002, September 2025 ↗
OWASP risk names are referenced for identification. Course scenarios and proposed mappings are sicurity.ai interpretations, not official OWASP materials or endorsements.