Irregular

Frontier AI models can be exploited for misuse — lateral movement, EDR evasion, ransomware-like behaviors — and existing testing is ad hoc and insufficient.

Irregular (formerly Pattern Labs) is the first dedicated frontier AI security lab. They build elaborate simulated environments where AI models are stress-tested as both attackers and victims under real-world adversarial scenarios.

Rapid deployment of powerful AI models by labs like OpenAI and Anthropic creates urgent demand for independent security validation before public release.

Already under contract with leading AI labs including OpenAI and Anthropic, plus government clients. Cited in safety evaluations for Claude 3.7 Sonnet, o3, and o4-mini. Valued at $450M.

Dan Lahav (CEO) and Omer Nevo (CTO), from Israel deep cybersecurity talent pool.