CalvinBackup/ai-llm-red-team-handbook
1
0
Fork 0
mirror of https://github.com/Shiva108/ai-llm-red-team-handbook.git synced 2026-08-27 21:30:27 +02:00
Code Issues Packages Projects Releases Wiki Activity

feat: Update visual recommendations in Visual_Recommendations_V2.md from Chapter 35 to Chapter 36 and add it to .gitignore.

Browse Source
This commit is contained in:
shiva108
2026-01-18 17:46:24 +01:00
parent a16c4fe9a4
commit c80d2c5fef
2 changed files with 42 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.gitignore
+2
View File
@@ -38,3 +38,5 @@ scripts/tests/verify_fixes.py
docs/Chapter_31_AI_System_Reconnaissance.md.backup
docs/Chapter_31_AI_System_Reconnaissance.md.audit_backup
docs/Visual_Recommendations.md
Visual_Recommendations_V2.md
.gitignore
Visual_Recommendations_V2.md
+40 -50
View File
@@ -1,59 +1,49 @@
# Visual Recommendations V2: Chapter 35 (Post-Exploitation)
# Visual Recommendations: Chapter 36 (Reporting and Communication)
## Analysis & Selection
## Style Constraints
**Chapter Focus:** Persistence, Lateral Movement, and Data Exfiltration.
**Selection Strategy:** Prioritize the flow of persistence (how an attack survives) and the "pivot" concept (how the LLM attacks other systems).
- **Aesthetic**: Deus Ex / Cyberpunk Renaissance
- **Colors**: Deep slate/black background. Accents in Gold (#C5A365) and Amber (#FFBF00).
- **Line Work**: Thin, crisp vector lines. Circuit board traces.
- **Typography**: OCR-B or technical monospace. No "Deus Ex" text.
- **Aspect Ratio**: 16:9 preferred.
---
## Visual Manifest
### VISUAL SPECIFICATION #1
### 1. Report Structure Pyramid
- **ID**: Ch35_Flow_Persistence
- **Target Section**: How Persistence Works
- **Context**: Illustrates the cycle of injecting a malicious prompt into the RAG database, which then gets retrieved and re-infects the model on subsequent queries.
- **Type**: Cyclic Flow Diagram
- **Elements**:
- Attacker (Injecting "Soft Prompt")
- Component: Vector DB (Storing the poison)
- User Query (Triggering retrieval)
- Retrieval Path (Poison -> Model)
- Outcome: Compromised Model Response
- **Style Notes**: Deus Ex aesthetic. Circular flow. The "Poison" should be a distinct red/gold data artifact infecting the blue/gold system.
- **Filename**: Ch35_Flow_Persistence.png
- **ID**: `Ch36_Structure_ReportPyramid`
- **Target Section**: 36.3 Structure of a High-Quality Red Team Report
- **Type**: Hierarchical Diagram
- **Description**: A stylized, technological pyramid or layered hologram.
- **Top Layer (Gold/Bright)**: Label "Executive Summary" (Context: Business Risk, ROI).
- **Middle Layer (Amber)**: Label "Findings Overview" (Context: Severity, Scope).
- **Bottom Layer (Darker/Slate)**: Label "Technical Evidence" (Context: Logs, PoC, Steps).
- **Style**: Floating holographic layers, connected by vertical data streams.
- **Filename**: `Ch36_Structure_ReportPyramid.png`
### VISUAL SPECIFICATION #2
### 2. Remediation Lifecycle Loop
- **ID**: Ch35_Concept_LateralMovement
- **Target Section**: Introduction (or Why This Matters)
- **Context**: Shows the LLM as a central hub (Brain) connected to external tools (Hands) like APIs, Databases, and Shell Access, illustrating lateral movement.
- **Type**: Network/Hub Diagram
- **Elements**:
- Central Node: Compromised LLM (The "Bridgehead")
- Connected Node A: Internal Database (SQL)
- Connected Node B: Cloud API (AWS/Azure)
- Connected Node C: Internal Network (Shell)
- Direction: Arrows moving FROM the LLM TO these systems.
- **Style Notes**: Deus Ex aesthetic. Central hub with radiating connection lines like a neural network or star topology. High-tech schematics.
- **Filename**: Ch35_Concept_LateralMovement.png
- **ID**: `Ch36_Flow_RemediationLoop`
- **Target Section**: 36.9 Deliverable Handoff and Follow-Up
- **Type**: Cyclic Process Flow
- **Description**: A circular feedback loop resembling a cybernetic engine or HUD interface.
- **Node 1**: "Report Delivery" (Icon: Document/Data transmission).
- **Node 2**: "Executive Buy-in" (Icon: Handshake or Shield).
- **Node 3**: "Re-Test / Verify" (Icon: Target reticle or scanner).
- **Center**: "Remediation" (Glowing core).
- **Style**: Rotating segments, gold data flow arrows, high-tech interface feel.
- **Filename**: `Ch36_Flow_RemediationLoop.png`
---
### 3. The Evidence Chain
## JSON Manifest
```json
[
{
"id": "Ch35_Flow_Persistence",
"filename": "Ch35_Flow_Persistence.png",
"description": "A cyclic technical diagram showing a 'Persistence Loop'. An Attacker executes an 'Injection' into a 'Vector Database' cube. A 'User Query' enters the loop, retrieves the 'Poisoned Data', and feeds it into the 'AI Model', which then outputs a 'Compromised Response'. Deus Ex style, dark background, gold and amber data paths, HUD elements.",
"tool_suggestion": "generate_image"
},
{
"id": "Ch35_Concept_LateralMovement",
"filename": "Ch35_Concept_LateralMovement.png",
"description": "A network diagram showing an AI Model as a central 'Pivot Point' connected to three targets: 'SQL Database', 'Cloud API', and 'Internal Shell'. Red attack vectors radiate from the central AI model to these targets. Deus Ex style, high-tech schematic, glowing connection lines, cyber-security aesthetic.",
"tool_suggestion": "generate_image"
}
]
```
- **ID**: `Ch36_Concept_EvidenceChain`
- **Target Section**: 36.4 Writing Style and Principles (or 36.7 Visuals)
- **Type**: Linear Data Pipeline
- **Description**: A left-to-right assembly line of data packets merging into a final "Finding".
- **Input 1**: "Raw Logs" (Binary stream).
- **Input 2**: "Screenshot" (Image frame).
- **Input 3**: "Exploit Script" (Code block icon).
- **Merge Point**: All feed into a central processor labeled "Validated Finding".
- **Style**: Data highway, assembly line, converging gold traces into a solid amber block.
- **Filename**: `Ch36_Concept_EvidenceChain.png`