# Fabraix Fabraix’s Nyx probes AI systems through adaptive, multi-turn adversarial tests to investigate security, logic and alignment weaknesses. - Findings24: - Site: - Preços: Freemium - Categorias: Assistentes de testes com IA ## O que é Fabraix? Fabraix provides Nyx, a long-running adversarial agent for testing AI systems. In an extended session, Nyx probes a target, observes its responses, adapts its next steps and follows promising leads to investigate security, logic and alignment weaknesses. Nyx needs a reachable endpoint, URL or phone number; it does not require source code, model weights or access to the target’s VPC. In a managed audit, Fabraix runs the assessment and returns an AIVSS-scored report with reproduction steps and remediation guidance. Testing boundaries are agreed in advance. Payloads are synthetic, secrets are canaries, and destructive or irreversible actions are observed and reported rather than executed against real assets. ## O que Fabraix pode fazer? ### Adaptive multi-turn assessments Nyx tests a target over an extended session, observes each response, adapts its next steps and follows promising leads instead of relying on a fixed set of checks. ### Black-box access Connect a reachable endpoint, URL or phone number. Nyx does not need source code, model weights or access to the target’s VPC. ### Coverage for different AI systems Nyx can test chatbots, autonomous agents, multi-agent systems, browser agents, voice agents and reinforcement-learning systems. ### Continuous retesting option Continuous is for teams retesting prompt, tool, and model changes before release. It is billed monthly by usage and lists multiple targets, CI/CD and scheduled scans, newly discovered attacks added continuously, a findings dashboard with export, regression tracking, and priority support. ## Casos de uso ### Check an agent before release Run a custom-priced One-Off Scan against one agent as a point-in-time adversarial audit before shipping. ### Retest a changing system Continuous is for teams that need to retest prompt, tool, and model changes before release. Its listed scope includes multiple targets, CI/CD and scheduled scans, and regression tracking over time; billing is monthly by usage. ### Run non-commercial AI research Academic and independent researchers can apply for free access to Nyx for non-commercial AI research; the listed researcher scope includes one target. ## Perguntas frequentes ### What is Nyx? Nyx is Fabraix’s long-running adversarial agent. It probes an AI system over an extended session, observes responses and adapts its next steps to investigate security, logic and alignment weaknesses. ### What kinds of systems can it test? Nyx can test chatbots, autonomous agents, multi-agent systems, browser agents, voice agents and reinforcement-learning systems. ### What access and setup does Nyx need? A target needs to be reachable through an endpoint, URL or phone number; the managed audit does not require source code, model weights, or access to the target’s VPC. Fabraix runs the managed audit and returns an AIVSS-scored report with reproduction and remediation guidance. Testing boundaries are agreed in advance; destructive or irreversible actions are observed and reported, not executed against real assets. ### How are test boundaries handled? Testing boundaries are agreed in advance. Payloads are synthetic and secrets are canaries; destructive or irreversible actions are observed and reported rather than executed against real assets. ### Which plan fits research or continuous testing? Researchers can apply for free non-commercial academic or independent AI research access for one target. A One-Off Scan is custom priced per target. Continuous is custom priced monthly by usage. ## Informações rápidas - **Software**: Nyx, provided by Fabraix - **Testing workflow**: Long-running, adaptive black-box assessments - **Target connection**: Endpoint, URL or phone number - **Setup**: Managed audit run by Fabraix; target reachable by endpoint, URL or phone - **Continuous plan**: Monthly by usage; multiple targets; CI/CD and scheduled scans; findings dashboard/export; regression tracking