The two founders come from elite Israeli military technology units, including Unit 81 and Unit 8200. Their company now works with some of the biggest names in artificial intelligence, including OpenAI, Anthropic and Meta, helping them test what their most advanced models can actually do.
Recent security tests showed just how serious the challenge has become. In several cases, AI models believed they were operating inside closed simulations but were able to reach real external networks because of mistakes in the testing setup. Some models then searched online, found weaknesses and interacted with real systems while still believing they were inside a controlled exercise.
Irregular builds the testing environments used to push these models to their limits and see whether they can follow rules, stay inside secure systems and avoid dangerous behavior. The company also works with governments, including the United Kingdom, and has helped develop more realistic security tests as AI models become capable of carrying out increasingly complex cyber tasks.
The startup raised about $80 million across two funding rounds led by Sequoia and Redpoint, with backing from major Israeli investors including Wiz co-founder Assaf Rappaport. It also became profitable in 2025 after signing large contracts with leading AI companies.
For Lahav and Nevo, the goal is bigger than building another cybersecurity company. They want Irregular to become a basic security layer for the AI era, helping companies like OpenAI, Meta and Anthropic make sure their most powerful models remain safe, controlled and inside the limits they were given.