◉ EMMANUEL TIGOUE

AI Security Engineer • SecurityX • SSCP • CCNA

I secure AI systems and build the governance around them. 14-container Zero Trust platform, 37 GRC documents, detection to alert in under 30 seconds.

Get In Touch View on GitHub

◉ CISSP IN PROGRESS // SPRING 2026

// Clearance Stack

Certifications

◉ IN PROGRESS

ISC² CISSP

Certified Information Systems Security Professional

Security Architecture & Engineering

Spring 2026

◉ VERIFIED

// Service Record

Experience

CoreDirective | AI Security Engineer

Atlanta, GA • Sep 2025 – Present

14

Container Stack

37

GRC Documents

80%

Triage Reduction

0

Exposed Ports

Secured AI gateway (Claude Opus 4.6, OWASP LLM Top 10, MITRE ATLAS) and red teamed 25 test cases for prompt injection, jailbreak, and data exfiltration pre-launch
Built shift-left CI/CD (Trivy, Semgrep, Gitleaks, OPA, Cosign, SBOM) and executed DAST with OWASP ZAP: zero injection vulns across 8 categories, 4 header fixes same-day
Cut alert noise from 200+ to 12 actionable findings by tuning Falco eBPF rules with Datadog routing via Falcosidekick
Defined 16 Terraform files, 30+ resources, 8 OPA policies. Deployed Teleport PAM (JIT) and Keycloak SSO (3-tier RBAC)
Reduced external attack surface to zero exposed ports through Cloudflare Zero Trust tunnels with mTLS
Assessed Google Cloud IAM across 7 APIs: OAuth 2.0 lifecycle, org-level policies, cross-domain identity federation
Automated ops via n8n SOAR with NeMo-sandboxed AI and Ollama local inference, cutting triage by 80%
Authored 37 GRC documents: SSP (133 NIST 800-53 controls), POA&M (27 findings), 10 policies, 5 IR playbooks, DAST assessment, pen test

Texaco | IT Security & Operations Manager

Atlanta, GA • Mar 2022 – Feb 2026

65%

Cost Reduction

90m

Containment Time

86%

Fewer Audit Findings

12h

Recovered Weekly

Led incident response across 3 locations: POS skimmer investigations, credential compromises, and vendor access incidents. Built 6-step IR runbook cutting containment from 8 hours to 90 minutes
Segmented flat network into 4 VLANs isolating POS payment, back-office, guest Wi-Fi, and management. Validated with Nmap scans
Deployed Splunk SIEM with correlation rules cutting mean time to detect from 48 hours to under 4 hours
Hardened Active Directory with GPO baselines, least-privilege enforcement, and credential rotation. Reduced audit findings from 14 to 2
Maintained PCI DSS compliance across 45+ devices with quarterly Nessus scans, segmentation validation, and SAQ documentation
Automated patch deployment, provisioning, and compliance reporting with Python and PowerShell, recovering 12 hours weekly
Established AI governance aligned to NIST AI RMF with LLM-powered phishing detection and incident prioritization

// Academic Record

Education

Georgia State University | Atlanta, GA

GPA: 3.7 • Dean's List

B.B.A. Computer Information Systems | May 2026
Concentration in Cybersecurity
B.B.A. Business Economics | May 2026
A.S. Business Administration | May 2025

// Proof of Work

What I Built

Operational security platform built on DigitalOcean, migrated from AWS EC2 for cost optimization. STRIDE threat-modeled, then deployed. Every metric below is verifiable in the public repository.

Deployed Containers

NIST Control Families

65%

Cost Reduction

OPA Policies

SOAR Workflows

<30s

Detection to Alert

GRC Documents

133

NIST Controls Mapped

IR Playbooks

        System Architecture // CoreDirective Security Platform
      

14 containers • 3 Docker networks • 7 layers • Click any layer to expand

▸Layer 1 // Perimeter & Edge

◉ Cloudflare Tunnel TLS 1.3

◉ Cloudflare DNS/WAF EDGE

Zero public ports exposed. All ingress through Cloudflare encrypted tunnels. WAF rules filter malicious traffic at the edge. DDoS protection and bot management at no additional cost. All services accessed through encrypted tunnel routes.

▸Layer 2 // Identity & Access

◉ Teleport PAM PAM

◉ Keycloak RBAC IAM

◉ Vault Secrets KMS

Teleport enforces JIT access. SSH sessions are time-limited, recorded, and auditable. Keycloak provides 3-tier RBAC with SSO. Vault manages dynamic credentials with automatic rotation. Zero hardcoded secrets in the entire stack.

▸Layer 3 // Docker Network Segmentation

◉ net-core BRIDGE

◉ net-ai AIR-GAP

◉ net-monitoring MONITOR

Three isolated Docker networks. net-core for service communication. net-ai is air-gapped (internal:true) so AI models cannot reach the internet at runtime. net-monitoring isolates the security stack from service traffic. n8n bridges net-core and net-ai. Event handler bridges net-core and net-monitoring.

▸Layer 4 // Core Services

◉ PostgreSQL 16 DB

◉ n8n SOAR SOAR

PostgreSQL stores workflow state with encrypted-at-rest storage and daily automated backups. n8n orchestrates 15 SOAR workflows across 16 integrated services: incident response, health monitoring, compliance checks, and operational intelligence.

▸Layer 5 // AI Pipeline (Air-Gapped)

◉ Ollama LLM

◉ Whisper STT STT

◉ OpenClaw AGENT

AI inference runs on an air-gapped network (net-ai, internal:true). Ollama and Whisper cannot reach the internet at runtime. OpenClaw gateway provides authenticated access to Claude Opus 4.6 with 6 active tool-use skills. Prompt injection defenses and output validation layers enforce AI safety.

▸Layer 6 // Detection & Response

◉ Falco eBPF

◉ Falcosidekick ROUTER

◉ Event Handler AUDIT

Falco monitors every syscall via eBPF with custom per-container rulesets. Detections route through Falcosidekick to Datadog dashboards and Telegram alerts in under 30 seconds. Event Handler captures all Teleport session recordings for compliance audit trails.

▸Layer 7 // Observability & Supply Chain

◉ Datadog APM

◉ Fluentd LOGS

◉ Trivy CVE

◉ Semgrep SAST

◉ Gitleaks SECRETS

◉ Cosign/Syft SBOM

Datadog provides full-stack observability with custom SOC dashboards. Fluentd aggregates container logs into a unified pipeline. CI/CD security gates: Trivy scans for CVEs, Semgrep performs SAST, Gitleaks prevents credential leaks, OPA/Rego enforces infrastructure policies, Cosign signs images, and Syft generates SBOMs for supply chain verification.

            DigitalOcean s-4vcpu-8gb • Ubuntu 24.04 • Terraform IaC • Docker Compose
            
Migrated from AWS EC2 (65% cost reduction) • Cloud-agnostic, redeployable in hours

// Cloud Migration Case Study

Problem

$135/mo

Single AWS EC2 instance. Vendor lock-in risk. NAT Gateway overhead. Underutilized compute.

Solution

Multi-Cloud IaC

Terraform-managed DigitalOcean infrastructure. Cloudflare zero-trust overlay. Full stack in Docker Compose. Cloud-agnostic architecture. Can redeploy to any provider in hours.

Result

$48/mo

Fraction of the cost. Zero downtime migration. Same stack, same security posture. Proves multi-cloud capability.

Zero Trust

4-Layer Security Architecture

STRIDE threat-modeled, then deployed. Cloudflare Tunnel (no exposed ports) + Teleport PAM with JIT access + Keycloak RBAC with 3-tier role separation. Falco eBPF runtime detection feeding Datadog SOC dashboards. Vault for secrets management.

Cost Engineering

65% Infrastructure Cost Reduction

AWS EC2 $135/mo down to DigitalOcean $48/mo. Self-hosted Qwen 3 8B handles local inference at zero marginal cost. Terraform IaC enables cloud-agnostic redeployment.

Automation

15-Workflow SOAR Orchestration

Single webhook-driven control plane integrating 16 services: Google Workspace, Telegram, PostgreSQL, Ollama, Cloudflare, GitHub, Gmail, and more.

TERRAFORM // compute.tf

resource "digitalocean_droplet" "cd_alpha" {
  name     = var.do_droplet_name
  region   = var.do_region
  size     = var.do_droplet_size
  image    = var.do_image
  vpc_uuid = digitalocean_vpc.default.id
  ssh_keys = [digitalocean_ssh_key.coredirective.id]
  tags     = var.do_tags

  lifecycle {
    prevent_destroy = true
    ignore_changes  = [image, ssh_keys]
  }
}

DOCKER // docker-compose.yaml

services:
  tunnel-cyber-squire:
    image: cloudflare/cloudflared:latest
    command: tunnel run
    network_mode: host
    restart: unless-stopped
    security_opt:
      - no-new-privileges:true
    read_only: true
    environment:
      - TUNNEL_TOKEN=${CD_TUNNEL_TOKEN}

  svc-automation:
    image: n8nio/n8n:latest
    security_opt:
      - no-new-privileges:true
    networks:
      - cd-internal # isolated bridge

OPA/REGO // enforce_vpc.rego

package terraform.digitalocean

deny[msg] {
  resource := input.resource_changes[_]
  resource.type == "digitalocean_droplet"
  not resource.change.after.vpc_uuid
  msg := "Droplet must be in a VPC"
}

deny[msg] {
  resource := input.resource_changes[_]
  resource.type == "digitalocean_droplet"
  not contains_tag(resource, "managed-by:terraform")
  msg := "All resources must be tagged"
}

// Security Engineering

Security Engineering

JUMP TO SUBSECTION ▼

▸ STRIDE Analysis ▸ Attack Paths ▸ Red Team ▸ Frameworks ▸ App Security

This is not generic threat modeling. This is adversarial analysis of an autonomous AI agent with tool-use capabilities: an LLM that can browse the web, execute code, query databases, trigger SOAR workflows, and push to GitHub. One compromised prompt can chain through 6 active skills into 16 downstream service integrations.

Every threat below was identified through STRIDE decomposition of the full agentic pipeline, then cross-referenced against MITRE ATT&CK/ATLAS, OWASP LLM Top 10, NIST 800-53, and ISO 42001. The result: 29 categorized threats, 4 mapped kill chains, and a clear picture of where the controls hold and where the gaps remain.

STRIDE Threats

AI Threats

Attack Paths

Trust Boundaries

30+

Data Flows

Frameworks

Agentic AI Threat Surface // Tool-Use Risk Map

Click any skill node to expand abuse scenarios and control mappings

CLAUDE OPUS 4.6 // AGENTIC GATEWAY

Autonomous AI agent with 6 active tool-use skills · Telegram-facing · No human-in-the-loop

Tavily Search

Real-time web search, content extraction

Indirect prompt injection via fetched content, data exfiltration through crafted queries

ChainSearch result → context poisoning → behavioral override

ATLASAML.T0051 (LLM Prompt Injection)

ControlOutput filtering, domain allowlist

Browser

Navigate websites, extract content, fill forms

SSRF to internal services, credential harvesting via phishing pages, JS execution

ChainBrowse → discover internal endpoints → exfiltrate data

ATLASAML.T0040 (ML Supply Chain Compromise)

ControlURL allowlisting, no localhost access, content sanitization

Python Interpreter

Execute arbitrary Python, file I/O, network calls

Remote code execution, filesystem traversal, reverse shell, crypto mining

ChainCode exec → filesystem access → credential theft → lateral movement

ATLASAML.T0043 (Craft Adversarial Data)

ControlSandboxed execution, no network by default, file path restrictions

GitHub

Create repos, push code, manage issues, read source

Malicious code injection, supply chain poisoning, secret extraction from repos

ChainRead source → identify secrets → push backdoored code

ATLASAML.T0042 (Verify Attack)

ControlToken scope limitation, branch protection, Gitleaks scanning

Notion

Create/read pages and databases, content management

Data exfiltration to external workspace, PII harvesting, social engineering content creation

ChainRead sensitive docs → exfiltrate via new page → cover tracks

ATLASAML.T0048.002 (Exfiltration via ML Inference API)

ControlWorkspace isolation, API token scope, audit logging

Gemini

Cross-model inference, content generation, analysis

Model-to-model prompt injection relay, response manipulation, oracle attacks

ChainQuery Gemini with poisoned context → use response to manipulate primary agent

ATLASAML.T0051 (LLM Prompt Injection)

ControlModel output validation, response filtering

COMPOUND RISK: 6 skills × 16 service integrations = 96 potential chain paths. A single prompt injection doesn’t compromise one tool. It compromises every tool the agent can reach. The attack surface isn’t additive. It’s multiplicative.

Interactive // STRIDE Threat Decomposition

Click any category to expand findings · 29 threats identified across 14 services

Spoofing

5 threats identified

Identity forgery across Telegram, webhooks, and API endpoints. Includes AI model identity spoofing.

Top ThreatTelegram identity spoofing via forged chat_id

AI ExtensionModel impersonation. Compromised gateway returning attacker-controlled responses

ControlsTelegram token auth, Cloudflare WAF, API key rotation, mTLS

NIST 800-53IA-2, IA-5, IA-8, SC-8

Risk LevelMEDIUM

Tampering

5 threats identified

Data and model integrity attacks targeting Docker images, database records, model weights, and audit logs.

Top ThreatDocker image manipulation via compromised registry pull

AI ExtensionOllama model weight alteration. Poisoned model producing backdoored outputs

ControlsCosign image signing, Trivy scanning, model hash verification, net-ai air gap

NIST 800-53SI-7, CM-3, CM-5, SA-12

Risk LevelMEDIUM

Repudiation

4 threats identified

Action denial and non-attribution in workflows, AI outputs, and audit trails.

Top ThreatAI-triggered workflow actions with no human attribution

AI ExtensionAI output non-attribution. No traceable link from AI decision to action taken

ControlsTeleport session recording, Fluentd log shipping, Datadog audit trails

NIST 800-53AU-2, AU-3, AU-6, AU-12

Risk LevelLOW

Information Disclosure

6 threats identified

Data leakage through AI inference, prompt exfiltration, credential exposure, and cross-zone data bleed.

Top ThreatPrompt leakage. Operational context sent to Anthropic API in inference requests

AI ExtensionPII in AI responses, system prompt extraction, side-channel inference attacks

ControlsPrompt sanitization, output filtering, local inference fallback, Vault secret isolation

NIST 800-53SC-8, SC-13, SC-28, SI-4

Risk LevelHIGH

Denial of Service

4 threats identified

Resource exhaustion targeting API budgets, GPU/CPU inference, container limits, and tunnel throughput.

Top ThreatAPI budget exhaustion. Adversary triggers expensive Claude API calls via crafted prompts

AI ExtensionContext window stuffing, model loading attacks against Ollama

ControlsRate limiting, API budget caps, container resource limits, Cloudflare DDoS protection

NIST 800-53SC-5, SI-4, CP-2, AU-6

Risk LevelMEDIUM

Elevation of Privilege

5 threats identified

Container escape, AI excessive agency, role escalation via Keycloak, Vault token abuse, n8n code execution.

Top ThreatAI excessive agency. Agent chains skills to execute unintended infrastructure actions

AI ExtensionPrompt injection → skill chaining → n8n workflow trigger → unauthorized DB/infra actions

ControlsSkill allowlisting, n8n workflow approval, Falco runtime detection, least privilege

NIST 800-53AC-6, CM-7, SI-3, SC-7

Risk LevelHIGH

Click any card to expand • Data sourced from THREAT_MODEL_STRIDE.md (629 lines)

Attack Tree // AI Pipeline Kill Chains

Path 1: Prompt Injection via External Messaging

HIGH PROB

Telegram→ Cloudflare Tunnel→ OpenClaw Gateway→ Override System Instructions

SKILL: LOW DETECT: MEDIUM OWASP LLM01 AML.T0051

Technique: Craft adversarial prompts to override system instructions or exploit multi-turn context drift. Includes encoding-based evasion (base64, unicode) and indirect injection via fetched web content (Tavily search results).
Controls: Input validation, system prompt hardening, behavioral monitoring via Datadog, conversation length limits, output validation before workflow triggers.
Gap: No automated prompt injection detection at the gateway level. Recommended: implement input classifier before LLM processing.

Path 2: Credential Theft via AI Agent

HIGH IMPACT

Prompt Injection→ Credential Disclosure→ n8n Workflows→ Lateral Movement

SKILL: MEDIUM DETECT: HIGH LLM06 AML.T0040

Technique: Chain prompt injection with credential extraction to trick the AI into revealing API keys, database credentials, or Vault tokens through its tool-use capabilities. Use exposed credentials to pivot to n8n workflows or database.
Controls: N8N_RESTRICT_ENVIRONMENT_VARIABLES_ACCESS=true (deployed), Vault dynamic secrets, credential scoping, Falco detection rules for unusual credential access patterns.
Mitigated: Environment variable restriction closes the primary credential theft vector through n8n.

Path 3: AI-Triggered Workflow Abuse

HIGH IMPACT

Excessive Agency→ Skill Chaining→ n8n SOAR Trigger→ Unauthorized Actions

SKILL: MEDIUM DETECT: MEDIUM LLM08 AML.T0040

Technique: Exploit the AI agent's tool-use capabilities to chain actions across browser, GitHub, and n8n. 16 service integrations in the SOAR platform create a wide action surface. Attacker crafts prompts that cause the AI to trigger destructive workflows.
Controls: Skill allowlisting, workflow-level approval gates, Falco detection for unexpected n8n execution patterns, action audit logging in PostgreSQL.
Gap: No human-in-the-loop gate for sensitive workflow triggers (database operations, infrastructure changes).

Path 4: Model Supply Chain Compromise

MED PROB

Ollama Registry→ Poisoned Model Pull→ Backdoored Inference→ Manipulated Outputs

SKILL: HIGH DETECT: LOW LLM03 AML.T0018

Technique: Compromise the Ollama model registry or intercept model pull to inject a backdoored model. GGUF format manipulation allows subtle behavioral changes that evade basic testing. No model signing by default in Ollama ecosystem.
Controls: net-ai air gap (no runtime internet), SHA256 hash verification at pull, periodic model integrity checks, behavioral baseline testing.
Gap: No automated model signing verification. Ollama registry lacks Sigstore/Cosign integration. Manual hash comparison only.

Full decomposition in ATTACK_TREE_AI_PIPELINE.md (344 lines)

Red Team Walkthrough // Prompt Injection → Skill Chain → Infrastructure Impact

Step through the highest-probability attack path. Click each phase to see detection status and control gaps.

Phase 01 // Reconnaissance

Target Identification

Attacker identifies public-facing AI agent with tool-use capabilities via messaging platform

Attacker discovers bot via platform search, sends benign queries to map capabilities, identifies skill set through response patterns. Passive reconnaissance reveals the agent’s available tools and behavioral boundaries.

PARTIAL // Platform metadata logged but no anomaly detection on reconnaissance patterns

▼

Phase 02 // Initial Access

Crafted Prompt Delivery

Adversarial prompt designed to override system instructions using encoding evasion techniques

Base64/unicode encoded prompt injection bypasses basic input filters. Multi-turn conversation builds trust context before injection payload. Technique: jailbreak + context manipulation.

PARTIAL // WAF active, no prompt-specific classifier at gateway

▼

Phase 03 // Execution

System Instruction Override

AI agent’s behavioral constraints bypassed. Attacker gains indirect control of tool-use capabilities

Agent now executes attacker-directed actions. System prompt integrity compromised. The agent believes it’s following legitimate instructions. No human-in-the-loop gate between prompt and action.

GAP // No runtime behavioral anomaly detection

▼

Phase 04 // Discovery

Internal Reconnaissance via Skill Chain

Compromised agent uses browser + GitHub skills to enumerate internal infrastructure and source code

Agent browses internal documentation, reads repositories for architecture details, identifies credentials in environment variables, maps orchestration workflow endpoints.

COVERED // Repository audit log + Falco runtime detection for unusual read patterns

▼

Phase 05 // Collection

Data Staging via Knowledge Base

Sensitive data aggregated into workspace pages for exfiltration staging

Agent creates new pages, copies credentials, infrastructure details, and PII from internal sources. Uses legitimate API. Looks like normal operations.

PARTIAL // API audit logged, no content classification on outbound data

▼

Phase 06 // Impact

SOAR Workflow Trigger

Agent triggers orchestration platform to execute unauthorized infrastructure actions

POST to master orchestrator webhook with valid format. 16 available service actions. Agent executes database queries, sends messages, modifies cloud infrastructure via API.

GAP // No human-in-the-loop approval for SOAR webhook triggers

▼

Phase 07 // Persistence

Backdoor via Code Push

Agent pushes modified workflow code to public repo, establishing persistent access independent of AI agent

Code management skill creates new branch, modifies CI/CD pipeline to include attacker payload, opens PR. If auto-merge enabled, backdoor persists even after AI agent is revoked.

COVERED // Branch protection + Gitleaks pre-commit + Cosign image signing

BLAST RADIUS:
16 service integrations compromised • Database access (PostgreSQL) • Cloud infrastructure (CDN, DNS) • Communication channels (messaging, email) • Code repositories • Knowledge bases

Framework Coverage Matrix

How each framework maps to STRIDE threat categories

NIST 800-53

OWASP LLM

MITRE ATLAS

NIST AI RMF

ISO 42001

CIS Benchmark

        Full Coverage
        Partial
        Not Applicable
      

Threat Modeling Deliverables

STRIDE Threat Model

29 threats • 14 services • 7 trust boundaries • AI extensions

View on GitHub →

Attack Tree // AI Pipeline

4 kill chains • MITRE ATLAS • OWASP LLM mapping

View on GitHub →

AI Threat Catalog

10 AI threats • 4 frameworks cross-referenced

View on GitHub →

Data Flow Diagram

30+ data flows • 7 trust boundaries • NIST SP 800-154

View on GitHub →

AI Supply Chain Risk Assessment

ML-BOM • Model provenance • Integrity verification

View on GitHub →

AI Red Team Plan

Adversarial testing • OWASP LLM Top 10 test cases

View on GitHub →

AI Incident Response Playbook

4 scenarios • Prompt injection • Excessive agency • Supply chain

View on GitHub →

View Full GRC Library on GitHub →

Application Security // Assessments & Pipeline

Vulnerability research, secure SDLC, code review, and IAM governance. Click any card for full documentation.

CI/CD Security Pipeline

PR ▶ fmt ▶ validate ▶ TFLint ▶ Checkov ▶ plan ▶ OPA ✓ 7 gates
MERGE ▶ Gitleaks ▶ Trivy ▶ Semgrep ▶ Cosign ▶ SBOM ✓ 5 gates

VULNERABILITY ASSESSMENT

SOAR Credential Exposure

CVSS 8.1 HIGH CWE-200

44 secrets exposed via process.env. Manual discovery. Remediated same day.

AC-6 OWASP A01 REMEDIATED

View on GitHub →

SECURE SDLC

CI/CD Security Pipeline

7 PR GATES 5 MERGE GATES

Gitleaks, Trivy, Semgrep, Checkov, OPA, Cosign, SBOM. 8 custom Rego policies.

SA-11 SA-15 CM-3

View on GitHub →

CODE REVIEW

Manual Security Assessment

1 HIGH 3 MEDIUM 1 LOW

5 findings from manual infrastructure code review. Logic flaws no scanner catches.

SA-11 CM-4

View on GitHub →

DAST METHODOLOGY

Dynamic Application Security Testing

OWASP ZAP Testing Guide v4.2

Zero exposed ports. All traffic through Cloudflare Tunnel. Quarterly assessment cadence.

RA-5 CA-8

View on GitHub →

CLOUD IAM ASSESSMENT

Google Cloud IAM Governance

OAuth 2.0 7 APIs Org Policy

Least-privilege IAM, credential lifecycle, cross-domain identity federation, org policy governance.

AC-2 AC-6 IA-2

View on GitHub →

OWASP TOP 10 (2025)

A01: Broken Access Control✓

A02: Security Misconfiguration✓

A03: Supply Chain Failures✓

A04: Insecure Design✓

A05: Injection✓

A06: Vulnerable Components✓

A07: Auth Failures✓

A08: Data Integrity Failures✓

A09: Logging & Alerting✓

A10: Exceptional Conditions✓

// AI Security & Governance

AI Security & Governance

Aligned to three frameworks: ISO 42001 · ISO 27701 · NIST AI RMF

AI Governance Policy aligned to three international frameworks, governing 3 deployed AI systems with documented risk profiles, prompt injection defenses, and behavioral monitoring.

Framework 1

ISO 42001

AI Management System

AI system inventory, risk classification, responsible AI principles, lifecycle management, human oversight requirements, and continuous monitoring controls.

View Policy on GitHub →

Framework 2

ISO 27701

Privacy Information Management

Data protection in AI pipelines, PII handling controls, consent management, data retention policies, privacy impact assessments for AI systems.

View Policy on GitHub →

Framework 3

NIST AI RMF

AI Risk Management Framework

Govern, Map, Measure, Manage functions applied to 3 deployed AI systems. Risk tiering, bias evaluation, performance monitoring, and incident response procedures.

View Policy on GitHub →

Deployed AI System Inventory

Ollama / Qwen 3 8B

Risk: LOW // Local inference

Self-hosted LLM. No data leaves server. Quantized for efficiency. Used for security analysis and automation tasks.

OpenClaw / Claude Opus 4.6

Risk: MEDIUM // External API

Authenticated gateway. Rate-limited. API key rotation enforced. Prompt injection defenses. Output validation layer.

Whisper STT

Risk: LOW // Local processing

On-premise voice transcription. Privacy-first: no audio leaves the server. No PII retention. Used for content workflows.

        Key Controls: Prompt injection defense • Output validation • Behavioral monitoring • Data retention policies • Human oversight requirements
      

// Runtime Detection // Falco Custom Rule

falco-rules/n8n-outbound.yaml PRODUCTION RULE

- rule: Unexpected outbound connection from n8n
  desc: Detects n8n container making connections to non-whitelisted IPs
  condition: >
    container.name = "svc-automation" and
    evt.type in (connect) and
    fd.typechar = "4" and
    not fd.rip in (rfc_1918_addresses)
  output: >
    Suspicious outbound connection from n8n
    (connection=%fd.name container=%container.name image=%container.image.repository)
  priority: WARNING
  tags: [network, n8n, lateral_movement]

Rule Breakdown

condition Scopes the rule to the svc-automation container only. Prevents false positives from other services. Monitors connect() syscalls caught by Falco eBPF.

fd.typechar Filters for IPv4 connections only ("4"). Combined with not fd.rip in rfc_1918_addresses. Any connection leaving the private network space triggers the alert.

output Structured output captures connection tuple, container name, and image. Shipped through Falcosidekick to Datadog dashboards and Telegram alerts for real-time SOC visibility.

tags Tagged lateral_movement. Aligns with MITRE ATT&CK T1571 (Non-Standard Port) and T1041 (Exfiltration Over C2). Enables automated correlation in Datadog SIEM.

Interactive // AI Deployment Decision Tree

Based on the AI Governance Policy (ISO 42001 • ISO 27701 • NIST AI RMF)

Does the AI system process, store, or have access to personal data (PII)?

// Governance, Risk & Compliance

GRC at a Glance

37 Documents

~20,000 lines of compliance documentation

NIST SP 800-53 MAPPED • 133 controls / 16 families

CIS Controls MAPPED • Risk Register

ISO 42001 / 27701 MAPPED • AI Governance

NIST AI RMF MAPPED • Risk Management

Self-assessed against NIST SP 800-53 Rev 5 Moderate baseline

▶ View GRC Library on GitHub →

Control Implementation Status

Implementation Coverage

Identity & Access92%

Teleport + Keycloak + Vault

Network Security95%

Cloudflare Tunnel, zero exposed ports

Detection & Response88%

Falco + Datadog + n8n SOAR (<30s detection-to-alert)

Data Protection85%

Encryption at rest, local AI processing

Compliance90%

37 GRC docs, 133 NIST controls

CI/CD Security94%

Gitleaks + Trivy + Semgrep + Cosign

NIST 800-53 Coverage by Family

Access Control

Audit & Accountability

Identification & Auth

Incident Response

Config Management

Risk Assessment

System & Comms

System & Info Integrity

Assessment & Authorization

Contingency Planning

System & Services Acq

Planning

Physical & Environmental

Personnel Security

Media Protection

Maintenance

High (>80%) Medium (40–80%) Low (<40%)

Key Deliverables

System Security Plan (SSP)

NIST 800-53 • 133 controls • 16 families

View on GitHub →

POA&M + Risk Assessment

27 entries • 17 threats • MITRE ATT&CK mapped

View on GitHub →

10 Security Policies

Including AI Governance (ISO 42001 + 27701 + NIST AI RMF)

View on GitHub →

5 IR Playbooks + Tabletop Exercise

Compromised container, DDoS, leaked credential, unauthorized access, AI system compromise

View on GitHub →

8 OPA/Rego Policies

Infrastructure enforcement • CI pipeline • Pre-commit hooks

View on GitHub →

IR Playbook // Compromised Container

TRIGGER

Container anomaly detected by Falco

Falco eBPF detects syscall anomaly or unexpected outbound connection. Alert routed via Falcosidekick to Datadog and Telegram within seconds. Incident timer starts.

▼

DECISION

Is the container internet-facing?

Check Cloudflare Tunnel routing table and Docker network config. Internet-facing: any container with tunnel ingress or exposed port. Non-internet-facing: internal bridge network only.

▼

YES // Isolate

ACTION

Isolate immediately (network disconnect)

Docker network disconnect to cut external access. Block Cloudflare tunnel route. Preserve container state. Do NOT stop or remove. Log isolation timestamp for SLA tracking.

NO // Assess

ACTION

Assess blast radius

Review Docker network topology. Identify containers on same bridge network. Check shared volumes. Assess if internal service credentials are in scope. Correlate with Datadog alerts.

▼

FORENSICS

Capture snapshot + analyze Falco audit logs

Docker export for filesystem snapshot. Teleport session recordings capture any interactive access. Pull Falco audit trail from Datadog. Preserve evidence before remediation.

▼

REMEDIATION

Rebuild from clean image

Pull fresh image from registry. Redeploy via Docker Compose with clean volume mount. Rotate any secrets that may have been exposed (Vault credential rotation). Verify all services healthy.

▼

Update detection rules • Document lessons learned

Post-incident review within 48 hours. Update Falco rules to close detection gap. Document in POA&M. File entry in GRC risk register. Close incident ticket with RCA summary.

Click any node to expand details

// Connect

Contact

Preferred

etigoue@tigouetheory.com

Response time: under 24 hours

Professional Networks

LINKEDIN linkedin.com/in/emmanuel-tigoue GITHUB github.com/ET-sec PORTFOLIO et-sec.github.io/portfolio RESUME Download PDF

Current Status

OPEN TO OPPORTUNITIES

LOCATION

Atlanta, GA

Open to relocation

CLEARANCE

Eligible

Download Resume

◉ EMMANUEL TIGOUE

Certifications

ISC2 CISSP

CompTIA SecurityX

Cisco CCNA

ISC2 SSCP

CompTIA Security+

CompTIA Network+

ISC2 CC

Experience

Education

What I Built

$135/mo

Multi-Cloud IaC

$48/mo

4-Layer Security Architecture

65% Infrastructure Cost Reduction

15-Workflow SOAR Orchestration

Security Engineering

AI Security & Governance

ISO 42001

ISO 27701

NIST AI RMF

GRC at a Glance

Contact

ISC² CISSP

ISC² SSCP

ISC² CC