Cinder Security
AI Security for AI-Native Companies
click to ignite
Research HTB Advisories Fracture Services Clients Pricing Contact
AI Security for AI-Native Companies

Secure your AI systems
before attackers adapt.

Authorized AI red teaming, RAG security testing, agentic workflow audits, and model supply-chain assessments for companies shipping production AI.

Request an Assessment View Research ↗
Public GitHub Security Advisories
HTB GCSB 2026 — AI Track 3/3
Responsible Disclosure
Stripe-ready Client Onboarding
GHSA-m4rw-22q2-87j8 — ModelEngine AI Tooling Advisory GHSA-4fpw-hjmg-x4qr — LangGraph / LangChain RAG Poisoning HTB GCSB 2026 — Project Nightfall AI Track Completed 3/3 Fracture — Autonomous AI Red Team Engine Authorized AI Security Assessments Stripe-ready Engagement Onboarding Responsible Disclosure — Safety-Bounded Testing GHSA-m4rw-22q2-87j8 — ModelEngine AI Tooling Advisory GHSA-4fpw-hjmg-x4qr — LangGraph / LangChain RAG Poisoning HTB GCSB 2026 — Project Nightfall AI Track Completed 3/3 Fracture — Autonomous AI Red Team Engine Authorized AI Security Assessments Stripe-ready Engagement Onboarding Responsible Disclosure — Safety-Bounded Testing

AI is shipping fast.
Audit capacity is not.

AI-native teams are deploying agents, RAG systems, memory layers, and model workflows at startup speed. Traditional security reviews rarely test the full trajectory: retrieval → tool call → memory update → action.

2
Public GitHub Security Advisories tied to Cinder Security research
3/3
HTB GCSB 2026 Project Nightfall AI track completed under authorized conditions
5
Core audit capabilities: trajectory, observation, memory, reference code, disclosure translation
6
Safety constraints: authorized, scoped, minimizing, auditable, reviewable, fail-safe
LATAM
Regional focus for AI security evidence, audit standards, and responsible deployment

Research that translates into controls.

Cinder Security turns adversarial evidence into audit-ready findings, executive reports, and remediation guidance for teams deploying AI systems.

0
Public GitHub Security Advisories
AI tooling + RAG orchestration
0
Controlled AI Challenges Solved
HTB GCSB 2026
0
AI Audit Capability Areas
Framework-driven
0
Safety-Bounded Testing Rules
Authorized only
0
Vendor Security Acknowledgment
0
Regional Mission
Latin America

Hack The Box GCSB 2026: AI Track Completed.

In May 2026, Cinder Security completed the AI challenge track in Hack The Box Global Cyber Skills Benchmark CTF 2026: Project Nightfall. These results are presented as controlled-environment evidence of technique viability, not as production vulnerability claims.

HTB GCSB 2026
Lotus Registry
ML supply-chain and model registry risk. Demonstrated how model artifacts, validation layers, and registry workflows can become trust boundaries.
Controlled✓ Solved
HTB GCSB 2026
Espionage Intelligence
RAG broken access control and compositional escalation. Demonstrated how retrieval weakness can compose with credential exposure and downstream execution paths.
Controlled✓ Solved
HTB GCSB 2026
Bribery Compliance
Agentic tool-result spoofing. Demonstrated how agents can reason faithfully over fabricated tool observations when observation authenticity is not enforced.
Controlled✓ Solved

Controlled environment · Authorized testing · No production claims · Technique viability only

Vulnerabilities we've disclosed.

Public, vendor-visible AI security research handled through responsible disclosure. Operational exploit payloads and unsafe reproduction details are intentionally omitted from public materials.

CSR-2026-002
ModelEngine / fit-framework
SSRF and prompt-injection composition in AI tooling. Demonstrates how model-mediated tool use can become an infrastructure security boundary when HTTP tools are exposed without sufficient filtering.
Critical ✓ Patch live
GHSA-m4rw-22q2-87j8 ↗ 🛡️ Vendor Acknowledgment ↗
CSR-2026-007
LangGraph / LangChain
RAG poisoning and indirect prompt-injection risk in orchestration. Demonstrates how retrieved context can corrupt agent trajectories even when the system behaves as designed.
High — CVSS 7.6 ✓ Public advisory
GHSA-4fpw-hjmg-x4qr ↗

Security Advisories.

Published security advisories and vendor-visible research from Cinder Security. Pending or non-public reports are excluded until they can be described accurately and safely.

AdvisoryCSR IDTargetTypeSeverityStatus
GHSA-m4rw-22q2-87j8 CSR-2026-002 ModelEngine fit-framework SSRF + Prompt Injection Critical Patch Live — v3.6.4 ↗
GHSA-4fpw-hjmg-x4qr CSR-2026-007 LangGraph / LangChain RAG Poisoning 7.6 High Public
🛡️ Vendor Acknowledgment
"We would like to thank Cinder Security (Esteban Ramos) for responsibly disclosing the critical SSRF vulnerability in the FEL LangChain plugin (CSR-2026-002)."
— ModelEngine Group, fit-framework v3.6.4 View Release ↗
MIT License

Fracture — autonomous AI red team engine.

An open-source CLI framework for safety-bounded offensive evaluation of AI systems. Fracture runs structured campaigns across RAG, agents, memory, tool use, and model workflows — producing evidence packages, severity mapping, and remediation guidance for authorized engagements.

cinder-security/fracture
MODULES
fingerprint
extract
memory
hpm
ssrf
retrieval_poison
obliteratus
campaign
shadow_replay
v1.0.0-cli · Phase 4 · Python · MIT
FRACTURE v1.0 — AUTONOMOUS AI RED TEAM ENGINE

Full-spectrum AI security assessments.

Authorized testing for AI-native teams shipping agents, RAG pipelines, model workflows, and generative AI features.

⚔️

AI Penetration Testing

One-time authorized assessment of your AI systems. Findings include reproducible evidence, severity mapping, and remediation guidance aligned to OWASP LLM Top 10.

One-time engagement
🔄

Continuous AI Red Teaming

Recurring safety-bounded testing as your AI evolves. Every model update, tool addition, and retrieval change can be evaluated before users encounter the failure mode.

Monthly retainer
🖼️

Diffusion Model Safety

Pre-launch safety evaluation for image generation models, including policy-bound prompt behavior, filter robustness, negative prompt handling, and adversarial-input resilience.

Pre-launch assessment
🎓

AI Security Training

Hands-on workshops for engineering and security teams covering AI threat modeling, RAG security, tool-use risks, memory-state safety, and responsible disclosure workflows.

Workshop
New

CinderGuard — your AI red team on autopilot.

24/7 autonomous red teaming for companies running AI agents in production. Fracture + CinderBot running continuous attack campaigns against your stack, surfacing vulnerabilities before they ship.

Continuous Campaigns

Automated attack campaigns run 24/7. New vectors tested on every deployment.

Real-Time Alerts

Instant notification when a new vulnerability is found. Severity-rated with PoC attached.

Monthly Executive Report

Board-ready security posture report. Trends, risk scores, and remediation progress tracked over time.

No Internal Team Required

Built for 50–200 employee companies without dedicated AI security teams. We are your team.

OWASP LLM Compliance

Every finding mapped to OWASP LLM Top 10. Audit-ready documentation included.

Shadow Replay

Re-run historical attack sessions against updated models. Verify your fixes hold.

$2,500 – $4,500/mo
Based on number of agents and attack surface scope
Request CinderGuard Demo Engagement Options ↗

Built for AI-native teams shipping fast.

Cinder Security works with AI startups and product teams deploying agents, RAG pipelines, model workflows, and generative AI features. Engagements are scoped privately, tested under explicit authorization, and delivered with executive-ready reporting.

AI-native startups
Authorized assessments
Stripe-ready onboarding
Executive reports

Transparent pricing.
No surprises.

Clear engagements with clear deliverables. Every assessment includes a professional report and debrief.

Starter
$750
USD — one-time
AI Security Assessment
  • Up to 3 attack vectors tested
  • Reproducible PoC per finding
  • Professional PDF report
  • 30-min debrief call
  • Delivered in 5 business days
  • 50% upfront · 50% on delivery
Request Scope
Recurring
CinderGuard
$2,500
USD/month — starting at
Autonomous 24/7 Red Team
  • Continuous Fracture campaigns
  • Real-time vulnerability alerts
  • Monthly executive report
  • Shadow replay on every update
  • Dedicated CinderBot agent
  • No long-term commitment
Request Demo
Stripe-ready payments
Request Invoice Request Stripe Payment Link ↗

Payments are accepted only after written scope approval. All testing is authorized, defensive, and contract-bound. Stripe · bank transfer · NDA available.

What we test.

The full attack surface of modern AI systems — from prompt-level exploits to infrastructure-level compromises.

Direct & Indirect Prompt Injection
Multi-turn Jailbreak Attacks
Psychological Manipulation (HPM)
System Prompt Extraction
RAG Pipeline Poisoning
Fine-tuning Backdoors & Data Poisoning
SSRF via AI Agent Tool Abuse
Tool & Function Call Hijacking
Multi-Agent Attack Chains
Model & API Key Extraction
Memory Poisoning in Persistent Agents
Guardrail & Safety Filter Bypass
MCP Server Injection
Diffusion Model Safety Evasion
Embedding-Space Adversarial Attacks

How we work.

A structured approach to finding what others miss.

01

Scope & Profile

Map your AI stack, identify attack surfaces, and define engagement rules.

02

Test & Validate

Run safety-bounded adversarial tests across agreed vectors. Every finding includes evidence and business impact.

03

Report & Remediate

Detailed security report with severity ratings, AI-specific risk mapping, and fix recommendations.

04

Verify & Harden

Re-test after fixes. Confirm vulnerabilities are resolved and defenses hold.

Ready to secure your AI stack?

Send us your scope. We will help you turn AI risk into a clear, authorized assessment plan.

contact@cindersecurity.io