Fabraix opens public access for Nyx AI security red team agent
Nyx is being offered as a blackbox, repeatable red-team system for customer-facing AI agents.
Fabraix says the agent can attack chat, voice, browser, and coding agents from only an endpoint, URL, or phone number. It builds test plans from YAML-defined audits, runs scheduled adversarial tests, and returns replayable findings with steps, responses, and failures. The company claims Nyx uses more than 10,000 jailbreaks and attack strategies, and reports a 78 percent success rate on AgentHarm. Access is through demo request, with CLI, REST API, and CI authentication support; pricing is not public. TestingCatalog's note
Fabraix says the agent can attack chat, voice, browser, and coding agents from only an endpoint, URL, or phone number. It builds test plans from YAML-defined audits, runs scheduled adversarial tests, and returns replayable findings with steps, responses, and failures. The company claims Nyx uses more than 10,000 jailbreaks and attack strategies, and reports a 78 percent success rate on AgentHarm. Access is through demo request, with CLI, REST API, and CI authentication support; pricing is not public. TestingCatalog's note
score 5