Irregular

Irregular, formerly known as Pattern Labs, describes itself as the first frontier security lab, with a stated mission of protecting the world as AI systems become more capable and sophisticated. It builds defenses through high-fidelity research platforms that simulate and monitor real-world AI security scenarios, aiming to identify emerging threats and vulnerabilities before models are publicly deployed.

The company's core work is running controlled simulations that expose advanced AI models to realistic threat scenarios, including antivirus evasion, mapping and analysis of system environments, and autonomous offensive cyber operations. It also assesses model resilience against targeted attacks and works on confidential AI inference and hardware-based isolation/verification mechanisms intended to protect data privacy and model weights. Outputs include evaluations used in model releases and frameworks such as SOLVE, a scoring system for a model's vulnerability-detection ability that is used within the industry.

Irregular publishes research on its website, including assessments of models against offensive security benchmarks and commentary on the direction of AI security. It was recruiting across AI research, cyber research, research and software engineering, security engineering, and technical policy as of the material reviewed.

Founding story

Irregular was founded in 2023 by Dan Lahav (CEO) and Omer Nevo (CTO). Lahav previously worked at LabPixies, a startup acquired by Google, and was an AI researcher at IBM, where he received the Outstanding Technical Achievement Award and published research including a cover feature in Nature. Nevo is a serial entrepreneur who served as a development manager at Google Research, leading AI projects such as wildfire detection models. The two met through competitive debate; Nevo is a world debate champion and Lahav holds the highest personal ranking in world championship history.

Business model

Irregular operates as a research-driven security lab rather than a seller of consumer-facing products, earning revenue from contracted evaluation and research work with AI developers and government bodies. It reports millions of dollars in annual revenue from clients including OpenAI and Anthropic, with research collaborations conducted directly alongside those labs.

Fees from contracted research, model evaluation and red-teaming engagements with AI labs and government clients.

Traction

Reported millions of dollars in annual revenue by September 2025, contracts with OpenAI, Anthropic and government clients, roughly 25 employees at the time of its funding announcement, and widespread citation of its evaluations and SOLVE framework across industry system cards and government use.

Latest developments

In September 2025 Irregular announced an $80 million round led by Sequoia Capital and Redpoint Ventures, reported at a $450 million valuation, and emerged from stealth under its new name. Funding is intended to support work on shaping global AI security standards and scaling research into commercially usable security capabilities. Its website lists publications dated August 2026, including an assessment of Kimi K3 against offensive security benchmarks and a post addressing recent incidents.