Nyx AI Red Team Agent: Fabraix Opens Public Access

Fabraix has opened public access to Nyx, an autonomous red-teaming agent designed for continuous security testing of customer-facing AI. What's interesting is that Nyx uses over 10,000 jailbreaks and attack strategies to test chat, voice, browser, and coding agents. It can target an endpoint, URL, or phone number and then configures its own test plan. Nyx operates as a pure blackbox, meaning it needs no source code or credentials. Fabraix reports a 78 percent attack success rate on a security benchmark called AgentHarm. The system provides detailed findings, including attack steps and agent responses, so tests can be replayed. Fabraix says Nyx often finds a first vulnerability within minutes or hours. The company also maintains a public Playground for the community to test live agents. The bottom line is that this tool aims to help security leaders and red teams protect AI agents used by customers.