Frontier evaluations
Build realistic security challenges and the tools to measure model performance. Study capability elicitation, generalization, and coverage to understand what models can reliably accomplish.
About
We evaluate frontier AI and build stronger cyber defenses.
Who we are
General Analysis is a frontier security lab. We study what advanced AI systems can do, how their capabilities can be misused, and how they can help defend critical systems.
We build realistic evaluation environments, agent frameworks, and training infrastructure for difficult security tasks. Our work measures frontier capabilities, investigates emerging risks, and develops more capable adversarial agents. We bring that research directly to enterprises through security testing and runtime protection for the AI systems they deploy.
What we work on
Training and adversarial research
Develop training environments and use reinforcement learning to improve agents for complex adversarial simulations. Investigate emerging capabilities and methods for mitigating their risks.
Security in deployment
Apply our research to the systems organizations deploy. Test agents against realistic attacks, identify weaknesses across tools and permissions, and protect their actions at runtime.
Team
Our founding team has published AI safety and security research in the world's most prestigious venues; worked on post-training at DeepMind, pretraining at Jane Street; and built the first safety models for policy enforcement at Cohere. More than half the team comes from Jane Street.
Get in touch