# Pentesting AI — Red Team, Blue Team & Security Testing | Utah

> Penetration testing AI and red-team/blue-team workspace. Model attack paths, audit OWASP Top 10 & APIs, and ship fixes from web or CLI.

Canonical URL: https://aithatsee.com/pentesting-ai
Workspace: https://aithatsee.com/utah/workspace?agent=attack#workspace

## Overview
Model real adversary paths with Attack, validate security posture, and immediately switch to Defense or Spartan to ship the remediation diff and regression suite.

## Key Specifications
- **OFFENSIVE & DEFENSIVE LOOP**: Attack + Defense + Spartan
- **MULTI-AGENT COORDINATION**: Shared Cloud Agent Rooms
- **BROWSER & LINUX EXECUTION**: Isolated Chrome & REA Lanes

## 01. Offensive hypothesis and defensive remediation in the same room.

Traditional penetration tests end in a static PDF weeks after code ships. Generic AI assistants refuse to reason past surface checklists. Utah lets authorized engineers test hypotheses, record verifiable evidence, and close the vulnerability in one continuous session.

- **Adversarial path & boundary modeling** (RED TEAM · ATTACK AGENT): Map authentication state machines, multi-tenant IDOR/BOLA vectors, OAuth redirect flows, GraphQL batching abuse, webhook signature bypasses, and privilege-escalation chains across your architecture.
- **Posture verification & detection engineering** (BLUE TEAM · DEFENSE AGENT): Verify TLS/HSTS, CSP frame-ancestors and script nonces, cookie flags, CORS origin reflection, rate-limit ceilings, and WAF/Sigma detections with clear separation between observed signals and hypotheses.
- **Production-ready patches without placeholders** (REMEDIATION · SPARTAN AGENT): Turn every confirmed finding into a complete code patch, strict input schema, and deterministic unit/integration test ready for pull-request review.

## 02. Structured coverage across modern web, API, and cloud stacks.

| Assessment Surface | What Utah Inspects & Models | Verification Output |
| --- | --- | --- |
| Authentication & Sessions | OAuth2/OIDC state/PKCE, JWT alg/aud/iss confusion, password reset race conditions, cookie scope | State-transition audit + hardened middleware patch |
| Multi-Tenant Authorization | Object-level IDOR/BOLA, role-claim spoofing, service-grant fencing, cross-tenant cache keys | Tenant-isolation matrix + negative test cases |
| Server-Side Request & Parsing | SSRF URL parsers, redirect hops, XXE/YAML/JSON prototype pollution, webhook replay windows | Allowlist/egress guard + deterministic parser tests |
| Client & Edge Headers | CSP script/style nonces, object-src, frame-ancestors, HSTS preload, COOP/CORP, security.txt | Live header posture receipt + exact Worker/Nginx config |

## 03. Drive red-team and blue-team reviews from your terminal or IDE.

### Run coordinated Chrome & repository checks from the Utah CLI
Switch seamlessly between local code inspection and isolated Defense/Attack browser lanes.
```bash
utah --browser-agent defense --task "/chrome Inspect https://example.com response headers and summarize CSP/HSTS gaps"
utah --mode deep --task "Review src/auth/ for token replay, timing leaks, and missing origin checks"
```

### Attach Utah 15-tool MCP to Cursor, Claude Code, or Codex
Write a 0600-permission MCP config with zero hardcoded secrets on disk.
```bash
utah --task "/ide cursor --write"
utah --task "/ide claude --write"
```

## Frequently Asked Questions

### Can I use Utah for penetration testing and red-team / blue-team exercises?
Yes. Utah provides dedicated Attack (authorized offensive research and threat modeling), Defense (posture hardening and detection), and Spartan (code remediation) workspaces, plus shared Agent Rooms where multiple AI sessions collaborate on an engagement.

### Does Utah distinguish between confirmed vulnerabilities and hypotheses?
Yes. Utah’s response protocol explicitly separates observed primary evidence (HTTP status, headers, code lines, SHA-256 receipts, test exit codes) from unverified hypotheses so reports never inflate an unconfirmed signal into a false claim of exploitation.
