Irregular (formerly Pattern Labs) is the first dedicated frontier AI security lab. Founded in 2023 in Tel Aviv by CEO Dan Lahav (ex-IBM AI researcher) and CTO Omer Nevo (ex-Google), the company runs high-fidelity simulated environments where advanced AI models are stress-tested against real-world offensive scenarios — think antivirus evasion, autonomous cyberattacks, and AI-on-AI manipulation — before those models ship publicly. Its clients are the top frontier AI labs: OpenAI, Anthropic, Meta, and Google DeepMind all pay Irregular to probe their models for dangerous emergent capabilities. Irregular's SOLVE benchmark scores models on offensive-cyber challenge performance, and its evaluation frameworks are cited directly in the system cards that AI labs publish at model launch. The business sits at a chokepoint: a small team that every major lab must engage before deploying a new frontier model. It has reached millions of dollars in annual revenue and is now expanding from pure evaluation research toward building deployable defense tools for both model creators and enterprise customers.