Fabraix is an AI research lab offering Nyx, an autonomous adversarial testing harness for AI agents. It probes security, logic, and alignment, surfacing failure modes manual tests miss, ensuring robust AI before deployment.
Freemium
Free (Research plan)
How to use Fabraix?
Users point Nyx at any AI system (e.g., chatbots, code agents, RL systems) with no special access required. Nyx runs thousands of parallel adversarial strategies across voice, text, and images, adapting in real time to find vulnerabilities. Integrate into CI/CD for continuous testing.
Fabraix 's Core Features
Adversarial Testing: Run 1,000+ offensive strategies probing security, logic, and alignment in real time, pure blackbox, adapting to your system.
RL Verification: Detect reward hacking before training runs, saving up to 30% in wasted compute on incorrect objectives in RL environments.
Multi-modal Interaction: Test across voice, text, and images, deploying test websites and custom files for comprehensive agent coverage.
Massively Parallel: Run thousands of interaction strategies concurrently, scaling coverage with compute, not human effort.
Continuous Coverage: Plug into CI/CD for re-testing every agent update, prompt change, or tool integration, ensuring ongoing security.
Blueprint Library: Industry-specific interaction blueprints for chatbots, code agents, RL systems, browser agents, and more.
Fabraix 's Use Cases
Security teams find vulnerabilities in AI chatbots before malicious users exploit them, ensuring safe customer interactions.
Developers catch broken refactors and unsafe code execution in coding agents with shell or git access.
Researchers probe reward hacking and sandbagging in RL systems, optimizing training efficiency.
Financial advisors stress-test for hallucinated advice and compliance gaps in automated financial advisory agents.
Healthcare teams probe clinical copilots for unsafe triage and PHI leakage, ensuring patient safety.
Fabraix 's Pricing
Research
$0
For AI Researchers. Access for academic and independent AI research. Limited capabilities. 1 target, core interaction library, basic scan coverage, standard findings report, community support.
Team
Custom
For Companies Building or Using AI. Full coverage for teams running adversarial testing in their SDLC. Multiple targets, full interaction library, uncapped scan depth, findings dashboard & export, priority support.
Enterprise
Custom
For Enterprises Deploying AI at Scale. Dedicated infrastructure, controlled access, advanced compliance controls. Unlimited targets, full Nyx testing suite, full remediation support, enterprise-grade security, SSO & audit logging.
