Inside General Analysis: The Startup Bringing Structure, Safety, and Clarity to Enterprise-Level AI Adoption

General Analysis was born from a simple observation shared by its co-founders, Rex Liu, Rez Havaei, and Alan Wu: organizations were racing to adopt AI technologies faster than they were learning how to secure them. The three founders, who began as a small research group focused on AI safety, saw growing momentum around AI tools across industries. "Many leaders don't yet know what the threat landscape looks like," Liu says. "Before we can solve problems, we need people to know what they are protecting against."

The company officially launched in the summer of 2024 and began building its full-stack platform early the following year. What started as purely research-driven work has evolved into an enterprise-grade security offering designed to help organizations implement AI with confidence. That evolution reflects a deeper philosophy the founders share: responsible AI must be built with the same discipline that shaped earlier generations of cybersecurity, even though the underlying technology behaves very differently.

Liu and Havaei met through poker circles, bonding first over game theory, strategy, and what Liu describes as "a shared obsession with understanding risk." Their backgrounds complement that mindset; Liu brings experience in penetration testing and offensive security, while Havaei has a strong foundation in AI safety research. Wu adds engineering depth and product vision, helping the team transform their safety research into usable, scalable solutions for enterprise clients. "The three of us approach problems from different angles, but with the same goal," Havaei says. "We want AI systems to be something people can trust."

That mission has translated into a four-part platform built around continuous testing and protection of AI assets. The first component is red teaming, the systematic process of probing AI systems to uncover vulnerabilities or policy violations. These exercises involve identifying edge cases in model behavior, stress-testing how systems interpret instructions, and finding ways an attacker might manipulate data or outputs. "Red teaming gives us a map of where things can go wrong," Liu explains. "It makes the invisible visible."

Their second component, guardrails, converts those findings into real-time protections. These guardrails classify and verify every input and output generated by an AI system, applying client-specific policies so that responses stay within safe and approved boundaries. The third component is AI asset management, which helps teams maintain visibility across their knowledge bases, models, agent pipelines, and data stores, ensuring nothing is overlooked as AI tools become embedded across the organization. The final piece, the GA standard, is the company's internal framework for responsible AI deployment. It aligns with broader compliance expectations while distilling them into practices that enterprises can apply day-to-day.

The company's suite of solutions extends across practical use cases that mirror how enterprises are deploying AI today: it supports internal employee agents, AI code assistants, developer copilots, healthcare-focused copilots, and knowledge-base-driven customer service agents, all underpinned by the same security framework.

According to Liu, one of the company's most notable discoveries came during a red-teaming exercise on a database-agent architecture. The team identified a vulnerability that could allow malicious instructions to manipulate internal tables, potentially exposing private data. Their proprietary system surfaced the issue and generated a fix, demonstrating how guardrails and red teaming work together. "It validated what we had been building toward," Wu says. "Security for AI requires proactive testing, not reactive patching."

This philosophy, anticipation over reaction, is a cornerstone of the company's differentiation. General Analysis aims to build and train proprietary models specifically to simulate a broad range of attacks, giving its team visibility into scenarios that enterprises may not yet know to look for. "Our goal is to stay ahead of the curve," Havaei says. "If we can understand tomorrow's risks today, we can help organizations adopt AI without hesitation."

While the technology itself plays a major role, the founders emphasize that AI security is also a trust-building exercise. Many enterprises want to deploy AI but are unsure how to evaluate its reliability or safety. By offering transparency, structured testing, and clear policy enforcement, the company hopes to reduce that uncertainty. "People want to use AI, but they want to know it's safe," Liu says. "Security gives them the confidence to move forward."

Looking ahead, General Analysis is focused on expanding its research team and strengthening its platform. The founders believe AI adoption will continue to accelerate, bringing with it new categories of risk that organizations must prepare for. Their long-term ambition is to help enterprises navigate that transition with clarity, combining research, security engineering, and thoughtful design to define what safe AI deployment can look like. Wu says, "Our responsibility is not just to protect systems today, but to help shape the standards that will guide them tomorrow."