General Analysis is a security research lab founded by Harvard, CMU, and DeepMind researchers helping enterprises adopt powerful agentic AI while being confident in their reliability and security. They build the models and infrastructure to threat-test and secure agentic systems — from adversarial evaluation and AI-specific threat intelligence to agent identity and runtime security controls. Much of their work is centered on understanding failure modes such as prompt injection, data leakage, multi-agent risk, and unsafe tool use.

Agentic AI is being deployed into consequential enterprise workflows faster than the security infrastructure to support it exists, and the prior generation of AI security tooling (built for deterministic LLM applications) is structurally unequipped to address it. GA's approach is empirical rather than policy-based: a proprietary RL-trained attacker model that generates context-aware exploits against live agent deployments. We believe this continuous research flywheel positions GA to become the security standard for the agentic era.