AI security firm Irregular announced Wednesday the completion of an $80 million funding round, co-led by Sequoia Capital and Redpoint Ventures. Wiz CEO Assaf Rappaport also participated in the round. A source familiar with the deal indicated the funding values Irregular at $450 million.
Formerly operating as Pattern Labs, Irregular specializes in AI evaluations, providing security assessments for frontier artificial intelligence models. The company's work has been cited in security evaluations for models such as Claude 3.7 Sonnet, as well as OpenAI's o3 and o4-mini. Its proprietary framework, SOLVE, designed for scoring a model's vulnerability-detection capabilities, is reportedly utilized across the industry.
Irregular's co-founder, Dan Lahav, stated to TechCrunch that "a lot of economic activity is going to come from human-on-AI interaction and AI-on-AI interaction," which he anticipates will "break the security stack along multiple points." The new capital is intended to support the company's efforts in identifying emergent risks and behaviors in AI models before they are deployed commercially.
To address these advanced threats, Irregular has developed an extensive system of simulated environments for intensive model testing. Co-founder Omer Nevo elaborated on their approach, noting, "We have complex network simulations where we have AI both taking the role of attacker and defender." This methodology allows the firm to assess a model's defensive strengths and weaknesses upon its release.
The emphasis on AI security has intensified across the industry as frontier models present new potential risks. OpenAI, for instance, overhauled its internal security protocols this past summer in response to concerns regarding corporate espionage. Concurrently, AI models themselves are demonstrating increasing proficiency in identifying software vulnerabilities, creating a dynamic landscape for both cyber defenders and potential attackers. Lahav acknowledged that securing these increasingly sophisticated models is a "moving target" requiring continuous effort.
