Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 10 additions & 8 deletions demos/ai-assisted-detection-demo/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -12,11 +12,13 @@ It combines deterministic detections with a tightly constrained LLM stage:

The LLM does **not** make final incident decisions, modify rules, call tools, or execute response actions. Human verification is always required.

## Purpose

The goal is to show a credible bridge between deterministic telemetry analytics and safe analyst assistance.

This is not an autonomous SOC. It is a constrained drafting pipeline that keeps rule logic, ATT&CK mapping, case grouping, and evidence handling deterministic.
## Purpose

The goal is to show a credible bridge between deterministic telemetry analytics and safe analyst assistance.

This is not an autonomous SOC. It is a constrained drafting pipeline that keeps rule logic, ATT&CK mapping, case grouping, and evidence handling deterministic.

For a no-run reviewer pack, see [docs/ai-assisted-detection-examples.md](../../docs/ai-assisted-detection-examples.md).

## Pipeline

Expand Down Expand Up @@ -93,9 +95,9 @@ The bundled sample data is designed to produce at least three generated cases.

Rejected outputs do not become analyst summaries. Analysts can still inspect deterministic evidence through `case_bundles.json`, `case_report.md`, and `audit_traces.jsonl`.

## Reviewer walkthrough
### Accepted summary path
## Reviewer walkthrough

### Accepted summary path

Use the default sample run artifacts in `artifacts/case_summaries.json`, `artifacts/case_report.md`, and `artifacts/audit_traces.jsonl`.

Expand Down
118 changes: 118 additions & 0 deletions docs/ai-assisted-detection-examples.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,118 @@
# AI-Assisted Detection Examples

This page is a small, sanitized examples pack for reviewers who want to inspect `ai-assisted-detection-demo` without running the pipeline.

The accepted path points to committed demo artifacts. The rejected and degraded-coverage paths use representative sanitized excerpts that match the fields emitted by the pipeline and exercised by tests.

## Accepted Summary Path

**Artifact files**

- `demos/ai-assisted-detection-demo/artifacts/case_summaries.json`
- `demos/ai-assisted-detection-demo/artifacts/case_report.md`
- `demos/ai-assisted-detection-demo/artifacts/audit_traces.jsonl`

**Fields to inspect**

- `case_summaries.json`: `case_id`, `human_verification`, `scope_guardrail`
- `case_report.md`: `Summary`, `Likely causes`, `Suggested next steps`
- `audit_traces.jsonl`: `validation_status`, `schema_version`, `rejection_reason`

```json
{
"case_id": "CASE-001",
"human_verification": "required",
"scope_guardrail": "no_final_incident_decision|no_rule_changes|no_automated_actions"
}
```

```json
{
"case_id": "CASE-001",
"schema_version": "ai-assisted-detection-audit/v1",
"validation_status": "accepted",
"rejection_reason": null
}
```

**What the reviewer should conclude**

- The accepted summary is bound to a real deterministic case and recorded in both analyst-facing and audit artifacts.
- The summary remains constrained draft output: `human_verification` is required and the scope guardrail forbids final verdicts, rule changes, and automated actions.

## Rejected Summary Path

**Artifact files**

- `case_summaries.json`
- `case_report.md`
- `audit_traces.jsonl`

**Fields to inspect**

- `case_summaries.json`: absence of the rejected `case_id`
- `case_report.md`: `Summary status`, `Rejection reason`, analyst note
- `audit_traces.jsonl`: `case_id`, `validation_status`, `rejection_reason`

Representative sanitized excerpt:

```json
{
"case_id": "CASE-EX-REJECTED",
"validation_status": "rejected",
"rejection_reason": "case_id_mismatch"
}
```

```md
## CASE-EX-REJECTED

Summary status: rejected
Rejection reason: case_id_mismatch
Analyst note: no accepted summary was produced for this case; rely on deterministic evidence.
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Show the correct rejected-path analyst note text

The rejected-path markdown example uses Analyst note: no accepted summary was produced..., but the runtime emits a different analyst note for rejected summaries (use the deterministic rule hits and raw evidence for manual review) when validation_status is rejected (see src/telemetry_window_demo/ai_assisted_detection_demo/pipeline.py around lines 940-948). Because this page is presented as a reviewer pack that matches pipeline output, the current text can mislead reviewers into expecting the wrong case report content when validating rejection behavior.

Useful? React with 👍 / 👎.

```

**What the reviewer should conclude**

- Invalid or disallowed model output is rejected fail-closed rather than patched or silently accepted.
- The rejection is auditable, and deterministic evidence remains available even when no accepted summary exists.

## Degraded Coverage Path

**Artifact files**

- `case_report.md`
- `audit_traces.jsonl`

**Fields to inspect**

- `case_report.md`: `accepted_rules`, `rejected_rules`, `coverage_degraded`, `rejection_reasons`
- `audit_traces.jsonl`: `case_id = null`, `rule_ids`, `rejection_reason`

Representative sanitized excerpt:

```md
## Run Integrity

- accepted_rules: AUTH-002, PROC-001, WEB-001
- rejected_rules: AUTH-001
- coverage_degraded: yes
- rejection_reasons: rule_metadata_validation_failed

Global validation rejections:
- AUTH-001: rule_metadata_validation_failed
```

```json
{
"case_id": null,
"rule_ids": ["AUTH-001"],
"validation_status": "rejected",
"rejection_reason": "rule_metadata_validation_failed"
}
```

**What the reviewer should conclude**

- Rule or ATT&CK metadata failures reduce coverage safely instead of causing a hard crash.
- The coverage loss is disclosed in the main report, not hidden only in the audit log.
Loading