Open-source · Built on weekends · Made with care

Red-team your
AI agents with ease.

Sixi AI is an open-source tool that automatically finds security vulnerabilities in your AI agents — chatbots, MCP servers, A2A endpoints — across 6 industry frameworks. Connect your agent, run a scan, get a clear report.

Built withAnthropic ClaudeGoogle GeminiOpenAI GPTLangGraph

19

ATTACK AGENTS

134

TECHNIQUES

6

FRAMEWORKS

3

LLM PROVIDERS

What It Does

Automated security testing for AI agents

Traditional security tools weren't built for systems that understand natural language. Sixi AI was designed specifically for the AI agent era.

Thinks Like an Attacker

19 autonomous agents probe for prompt injection, data leakage, excessive agency, and goal hijacking — the same vectors real adversaries use.

Clear, Actionable Reports

Every finding comes with evidence, severity scoring, framework mapping, and step-by-step remediation. No vague risk scores.

Any Agent, Any Protocol

REST chatbots, MCP tool servers, A2A agent networks — one tool to test them all, regardless of how they're deployed.

Framework-Mapped Results

Findings map to OWASP LLM Top 10, MITRE ATLAS, STRIDE, LINDDUN, MAESTRO, and PASTA. Great for audits and compliance reviews.

How It Works

Three steps — that's it

01

POINT IT AT YOUR AGENT

Connect to any AI agent — chatbot, MCP server, or A2A endpoint. REST, WebSocket, or protocol-native. Takes about a minute.

02

LET IT RUN

19 attack agents run 134 techniques in parallel with multi-model rotation. Grab a coffee — it handles the rest.

03

READ THE REPORT

Get severity-scored findings with evidence, framework mappings, and clear remediation steps. Export as PDF or JSON.

Use Cases

Useful anywhere AI talks to people

If you're building or deploying AI agents that interact with users, partners, or other AI systems — Sixi AI can help you find the weak spots.

Finance & Banking

AI advisors, fraud-detection bots, trading assistants — test them for prompt injection and data exfiltration before they go live.

Pharma & Life Sciences

Clinical data agents, drug-interaction checkers, patient-record bots — make sure they resist prompt attacks and don't leak sensitive data.

Healthcare

Triage chatbots, diagnostic assistants, patient portals — catch hallucinated medical advice and unauthorized data access early.

SaaS & Enterprise

Customer-facing chatbots, internal copilots, MCP tool servers — if users can talk to it, it should be tested.

Government & Public Sector

Citizen-facing AI assistants, policy chatbots, internal workflows — evidence-backed assessments for compliance requirements.

Retail & E-Commerce

Shopping assistants, recommendation engines, support bots — prevent goal hijacking and inventory exfiltration.

Framework Coverage

Mapped to six security frameworks

Every finding is mapped to the frameworks security teams already know — making reports easy to understand and act on.

OWASP LLM Top 10

Prompt injection, data leakage, excessive agency

MITRE ATLAS

Adversarial ML threat framework

MAESTRO

7-layer agentic AI reference model

STRIDE

Threat classification taxonomy

LINDDUN

Privacy threat modeling

PASTA

Risk-centric threat analysis

Curious what it finds?
Give it a try — it's free.

Sign up, point it at your agent, and see the results. No credit card, no strings attached.

About

A side project by friends
who care about AI security.

Sixi AI started as a weekend project between colleagues across Europe. We noticed that everyone is shipping AI agents — chatbots, MCP servers, A2A networks — but barely anyone is testing them for security vulnerabilities.

So we built a tool that does it automatically. It's open-source, it's free to try, and we keep improving it because we genuinely think this problem matters. If you find it useful, we'd love to hear from you.

The Team

Built across Europe

We're a small group of engineers and security folks spread across three countries, collaborating on evenings and weekends.

🇨🇭 Switzerland🇪🇸 Spain🇵🇱 Poland

Open source, privacy-first

Your scan data stays on your machine. No telemetry, no tracking. The tool runs locally — we never see your agent's data.