AI RED-TEAMING

Red-Team Your AI Agents

Burp Suite for AI agents — from 15 Research Lab

Multi-step attack chains mapped to MITRE ATLAS. Test whether your Authensor policies actually hold under adversarial pressure. Desktop app built in Rust and Tauri. MIT licensed.

WHAT IT DOES

Attack chains that prove your defenses.

Authensor tells your agents what they can and cannot do. Chainbreaker tests whether that actually holds. It is the offense layer of AI safety — a structured workbench for discovering how your policies break under adversarial conditions.

Multi-step attack chains mapped to MITRE ATLAS TTPs.

MCP security scanner catches tool poisoning and rug pull attacks before they reach your agent.

15-dimension Chainbreaker Score quantifies risk across behavioral and security axes.

Desktop app in Rust and Tauri. MIT licensed. No cloud required.

FEATURES

Six attack surfaces. One workbench.

Every component is purpose-built for AI adversarial testing. No generic security scanners repurposed for AI — Chainbreaker is built from the ground up for the agent threat model.

Multi-Step Attack Chains

Build and execute attack chains with visual chain editor. Every TTP is mapped to MITRE ATLAS so you know exactly what adversarial behavior you are testing.

ATLAS-mapped TTPsVisual chain builderChained payloads

MCP Security Scanner

Scan Model Context Protocol servers for tool poisoning, rug pull attacks, and schema manipulation before your agents connect to them.

Tool poisoningRug pull detectionSchema manipulation

Intruder Engine

Four attack modes borrowed from the professional security world: Sniper, Battering Ram, Pitchfork, and Cluster Bomb. Fuzz prompts, parameters, and tool inputs systematically.

SniperBattering RamPitchforkCluster Bomb

15-Dimension Scoring

The Chainbreaker Score combines behavioral and security dimensions into a single quantified risk number. Track improvements over time as you harden your policies.

Behavioral dimensionsSecurity dimensionsTime-series tracking

Canary Tokens

Plant canary tokens in your agent context and detect out-of-band exfiltration. Know immediately when data escapes through covert channels.

Out-of-band detectionDNS canariesExfiltration alerts

7 Report Formats

Export findings in the format your workflow needs: Terminal, JSON, Markdown, HTML, SARIF, EU AI Act, and NIST. Drop SARIF directly into your CI pipeline.

SARIFEU AI ActNISTHTMLMarkdown
THE LOOP

Build. Protect. Test. Improve. Ship.

Authensor and Chainbreaker are two sides of the same coin. Authensor is the defense. Chainbreaker is the offense. Run both on every iteration.

1
BuildDevelop your AI agent.
2
ProtectApply Authensor policies and content safety.
3
TestRun Chainbreaker attack chains against your defenses.
4
ImproveUse the Chainbreaker Score to close the gaps.
5
ShipDeploy with confidence. Repeat on every change.
Authensor

Defense — policies, approvals, content safety, audit trail.

Chainbreaker

Offense — attack chains, fuzzing, scoring, evidence.

BY THE NUMBERS

Built for thoroughness, not theater.

15
scoring dimensions
7
report formats
4
Intruder attack modes
MIT
open-source license

Find out if your defenses hold.

Chainbreaker is MIT licensed and built by 15 Research Lab. Run attack chains locally against your Authensor-protected agents.