Irregular

Irregular is a frontier AI security lab that conducts red teaming, safety evaluations, and misuse testing on advanced AI models. Its client list is about as credible as this space gets: OpenAI, Anthropic, and Google DeepMind, along with government agencies, use its platform to stress-test frontier models for cyberattack capability and unexpected behavior before and after deployment.

Founded in Tel Aviv in late 2023 by CEO Dan Lahav and CTO Omer Nevo, Irregular raised roughly $80 million across two rapid rounds in September 2025, led by Sequoia and Redpoint with participation from Wiz co-founder Assaf Rappaport and Eon founder Ofir Ehrlich among its investors, at a reported valuation near $450 million. The company became profitable during 2025 on the strength of contracts with major AI labs.

Irregular’s client roster is independently confirmed — not just claimed — because Anthropic itself publicly disclosed the relationship in 2026 while describing a serious incident: a misconfiguration in Irregular’s testing environment inadvertently connected sandboxed evaluation systems to the public internet, allowing AI models under test to access real-world systems during cybersecurity evaluations. Anthropic said the earliest incidents dated to April 2026, suspended all cyber model evaluations with Irregular on July 23, and began notifying other affected organizations on July 27. That’s a genuine, independently documented operational security lapse at the core of what Irregular sells — safe testing environments for dangerous capabilities — and it materially tempers what would otherwise be a very strong profile built on elite customers and funding.