Fabraix opens public access to Nyx, its AI security red-team agent. TestingCatalog reports that Nyx has more than 10,000 attack strategies and a 78 percent AgentHarm attack success rate. The report says Fabraix, founded in 2026 by Ahmed Aly and Ibrahim Abdu in the YC S26 batch, is among vendors finding exploitable agent failures at dozens of Fortune 500 companies.
Fabraix publishes Adversarial Cost to Exploit (ACE), a dynamic benchmark that measures AI agent security by the token spend an attacker must invest to breach an agent. Ibrahim Abdu and Ahmed (Zach) Aly author the benchmark, which scores safety according to the attacker's required token cost.
Fabraix Playground is an open-source platform for stress-testing live AI agents with real tools and published system prompts through adversarial play. It publishes winning jailbreak techniques to advance collective understanding of AI agent security.
