Summary
Fabraix launched publicly as an adversarial verification platform for AI agents, pairing black-box stress testing with runtime defense. The product is built around finding functional failures, security exploits, and autonomy gaps before production users trip over them.
What changed
Fabraix launched its public product and docs for adversarial verification and runtime defense for AI agents, including offensive testing and action-check flows.
Why it matters
Reliability and security are quickly becoming the real bottlenecks in agent deployment. Fabraix is part of the emerging class of products that treats adversarial testing, runtime checks, and exploit discovery as standard agent infrastructure instead of specialist security work.
Evidence excerpt
Fabraix says it uses offensive AI to stress-test agents and provides runtime checks to block prompt injection, goal deviation, memory poisoning, and unsafe actions.