Irregular is a frontier AI security research lab focused on understanding and addressing security risks posed by increasingly capable AI systems. The company conducts original research at the intersection of artificial intelligence and cybersecurity, with a particular emphasis on evaluating offensive capabilities of frontier AI models. Key activities include: - Publishing research on AI model behavior in offensive security contexts, such as assessing large language models (e.g., GPT-5.4-Thinking, Meta's Muse Spark) against offensive security benchmarks - Studying emergent cyber behaviors in AI agents, including scenarios where AI agents exhibit offensive threat actor characteristics - Advising or serving leading AI laboratories, indicating the company's work is relevant to AI developers seeking to understand the security implications of their systems Irregular's research focus spans AI safety and cybersecurity policy, as evidenced by roles including Technical Policy Researcher alongside AI Researchers and Cyber Researchers. The company is backed by investors and trusted by major AI labs, suggesting it operates primarily as a research and advisory entity rather than a traditional software product vendor. The team composition — AI researchers, cyber researchers, research engineers, and policy researchers — reflects a multidisciplinary approach to frontier AI security. Irregular does not appear to sell a commercial software product; its core output is security research and publications aimed at understanding the threat landscape introduced by advanced AI systems.