Irregular raises $80 million to secure frontier AI models

On Wednesday, the AI security startup Irregular announced a successful funding round, raising $80 million. This round was spearheaded by prominent investors Sequoia Capital and Redpoint Ventures, with contributions from Wiz CEO Assaf Rappaport. A source familiar with the funding disclosed that the valuation of Irregular has now reached $450 million. Co-founder Dan Lahav shared insights with TechCrunch, expressing that the future will see a surge in economic activities driven by interactions between humans and AI, as well as between AI systems themselves. He emphasized that this shift could potentially expose vulnerabilities across various security levels. Previously known as Pattern Labs, Irregular has established itself as a notable entity in the realm of AI evaluations. The firm’s contributions are referenced in security assessments for models such as Claude 3.7 Sonnet and OpenAI’s o3 and o4-mini versions. Their innovative framework, known as SOLVE, for evaluating a model's vulnerability detection capabilities, has gained traction within the industry. While Irregular has focused on assessing the current risks associated with AI models, their fundraising efforts are aimed at a more ambitious goal: preemptively identifying emerging risks and behaviors before they become problematic. The company has developed a sophisticated system of simulated environments to facilitate rigorous testing of AI models prior to their deployment. Co-founder Omer Nevo explained the process, stating, “We conduct complex network simulations where AI plays both the attacker and the defender roles. This allows us to identify the strengths and weaknesses of defenses when new models are introduced.” As the AI industry grapples with heightened security concerns linked to advanced models, OpenAI has revamped its internal security protocols this summer to counteract potential corporate espionage. The increasing capability of AI systems to discover software vulnerabilities poses significant challenges for both security professionals and attackers alike. For Irregular’s founders, this marks just the beginning of the security challenges posed by the advancements in large language models. Lahav articulated their mission succinctly: “As frontier labs strive to develop more sophisticated models, our mission is to secure them. However, with the evolving landscape, we recognize that there is an immense amount of work still ahead.”