General Analysis is Building the Security Arsenal for the Agentic Era

General Analysis, providing a comprehensive suite of AI safety tools, including run-time guardrails, red-teaming frameworks, interpretability techniques, observability, and more, led by Rez Havaei, Maximilian Li, and Rex Liu, has raised $10 million in seed funding led by Altos Ventures with participation from 645 Ventures, Menlo Ventures, Y Combinator, Leonis Capital, & strategic funds & angels.

Agent Security Is an Empirical Problem

Traditional software is deterministic and interpretable, making failures traceable to specific bugs and security largely a matter of architecture and access control. AI agents operate differently, handling complex, open-ended tasks that cannot be fully defined in code and therefore require broader, more flexible access.

At the same time, agents lack human accountability and do not offer strong guarantees in adversarial environments, making them inherently less robust. Treating them like either software or humans falls short.

This creates the need for a new security paradigm. Rather than relying solely on static guardrails, the focus shifts to reducing real-world risk through rigorous, technical countermeasures, moving beyond surface-level protections toward systems that can withstand evolving threats.

Innovation to Make AI Reliable and Secure

General Analysis was founded to address the growing challenge of agent reliability, bridging the gap between controlled simulations and real-world deployment. Backed by $10M in seed funding led by Altos Ventures, the company is building an enterprise platform focused on making AI agents secure and dependable.

The platform operates across two core areas. First, it offers a comprehensive suite of defensive tools, ranging from guardrails and prompt hardening to observability and identity management. Second, it runs adversarial simulations to test how agents perform under different failure scenarios, enabling tailored security configurations rather than one-size-fits-all solutions.

The broader goal is to help enterprises systematically improve robustness and safely deploy agents in higher-stakes environments, supported by ongoing research and proprietary models that stay ahead of emerging risks.

The Approach

General Analysis is focused on solving agent security at a fundamental level, not just certifying systems, but rigorously testing and improving them. While policy frameworks and governance have helped build trust, a critical gap remains: deep technical threat assessment and proactive risk mitigation.

The approach centers on empirically evaluating agent behavior, identifying real vulnerabilities, and strengthening systems through targeted defenses. Especially in high-stakes environments, meaningful adoption depends on transparent testing, honest risk evaluation, and a clear understanding of trade-offs.

The goal is not to slow deployment, but to build the confidence required to move forward responsibly.