“Our ambitions are huge. We know we are irregular in the landscape, hence the company’s name, but at the same time, we find ourselves sitting in conference rooms with the biggest AI players, and our contract with Anthropic bears the signature of founder Dario Amodei,” say Dan Lahav and Omer Nevo, founders of the startup Irregular, which they describe as “the world’s only security lab for advanced artificial intelligence.”
According to Lahav, the CEO, and Nevo, the CTO, “an extraordinary business opportunity has emerged to build a powerhouse focused on model control. After Anthropic executives told us, ‘You are worth the two hours a week that we spend with you,’ we realized the scale of the opportunity. That’s why we decided to raise capital, even though we unexpectedly became profitable in 2025.”
1 View gallery
Irregular founders.
(Photo: Yonatan Blum)
If Irregular’s plans come to fruition, many of today’s cybersecurity companies could become obsolete. The company is already working not only with Anthropic, but also with Google and OpenAI, as well as with governments, including the British government. Irregular is building tools that allow clients to assess how advanced AI models behave under threat and to develop methods to reduce risk.
For now, the signs suggest that Irregular is gaining traction. The funding Lahav and Nevo refer to, announced in September 2025, consisted of two rounds completed within weeks. The first round saw Sequoia Capital invest $30 million, followed by a second round of about $50 million, in which Sequoia participated again alongside Redpoint Ventures, Omri Caspi’s Swish Ventures, and several Israeli angel investors, including Wiz founder Assaf Rappaport and Ofir Ehrlich of Eon. Since then, the founders say, the company has been receiving constant inbound interest from investors. “We get proposals for funding every day, non-stop, and at times it’s no longer pleasant.”
On the surface, developing cybersecurity solutions for AI is one of the most crowded areas in the 2026 startup landscape. But Irregular argues its approach is fundamentally different. “The challenge lies in the existing cybersecurity mindset, where many conventions are now being broken due to changes in underlying infrastructure,” they say. “Before the internet, people kept money in safes guarded by humans. Then Check Point came along and redefined how information and value are protected. Today, we are in the middle of a similar shift, and Irregular aims to be the Check Point of the AI revolution.”
“To be as good at debating as we are, you have to listen”
Lahav and Nevo are not typical cybersecurity founders. Although both served in elite IDF intelligence units, they first met at the debating club at Tel Aviv University, where they excelled and each separately won world championships.
“Debating is one of the most enjoyable things you can do at university,” they say. “You’re given a topic and a position, and within 15 minutes you have to prepare a full argument.” Lahav recalls representing Iran’s position in a global competition on whether it should stop funding proxies in the Middle East. “It was fascinating to get inside that mindset.”
The skillset, they say, translates directly into business. “Our investors sometimes joke that ‘this isn’t a debate,’ but these skills are invaluable. We know how to disagree without taking it personally and how to quickly understand what the other side is really concerned about. To succeed in debating, you must listen carefully, and that applies to everything.”
Beyond debating, both founders were early technology enthusiasts.
“I’ve been obsessed with AI since I was six, when my father gave me a collection of Isaac Asimov stories,” says Lahav (36). “At 14, I was already working in tech during my summer holidays, and one startup I worked at was sold to Google for $20 million.” After serving in Unit 81, he pursued academic studies, including a master’s degree in bioinformatics.
Nevo (40) spent 12 years in Unit 8200 and helped found the Arazim program, which trains students in computer science and mathematics for intelligence roles. “I joined my unit without an academic degree, even though most of my peers had one,” he says. “Later, I helped design the academic track for the program, and eventually completed it myself.”
After leaving the military, Nevo founded NeoWize, which was later acquired by Il Makiage. He then joined Google, where he established a research team focused on detecting and analyzing wildfires, developing tools to predict their spread and improve response times.
“Something was missing, and it became a startup”
In 2023, both founders independently concluded that AI was approaching a critical inflection point. When they met, they realized that “something was missing in the world, and it took the form of a startup.”
“What we knew how to build didn’t yet exist,” they say. “And when you encounter that kind of opportunity, which may only come once in a lifetime, you drop everything and pursue it.”
Why did they believe they could offer something new?
“When we entered rooms with people shaping the future of AI, we were often the only ones who deeply understood cybersecurity at the level of advanced attacks,” says Nevo. “We’re not focused on what made headlines this week, but on what will appear in the future.”
Their entry into those circles, they admit, required a degree of boldness. “We reached out to contacts, explained that AI would create entirely new security risks, asked them what problems they couldn’t solve, and offered to tackle them for free. That’s how we ended up working with OpenAI and Anthropic, and even found ourselves discussing model defense concepts with Sam Altman.”
How do you define the problem Irregular is trying to solve?
“We operate as an applied lab for AI security, similar to how OpenAI functioned in its early days,” they explain. “In research we conducted, we showed that AI systems can manipulate many existing security mechanisms. In one case, two AI agents communicated with each other, recognized that safeguards were limiting their ability to complete a task, and devised a way to bypass those protections, effectively creating a coordinated workaround.”
For Lahav and Nevo, this represents a fundamental shift. “Cybersecurity is becoming a problem of controlling AI systems themselves. Traditional defenses were not designed to handle that.”
Translating that into a business model, they say, began with infrastructure. “Our first product allows us to analyze models, simulate attacks, and identify new types of vulnerabilities, which we then use to build the next generation of defenses. Over time, many of the people we initially worked with have become senior decision-makers, and last year we signed several seven-figure contracts. That’s what made us profitable in 2025, almost by accident.”
Do they believe other cybersecurity companies are approaching the problem incorrectly?
“We have a lot of respect for others in the field,” they say. “But we believe the core challenge is system control. The categories that define cybersecurity today will look very different in the future. Our ambition is to build a large, foundational company, similar to what emerged during previous infrastructure shifts with companies like Palo Alto Networks and CrowdStrike.”
Company ID
Sector: AI
Established: 2023
Founders: Dan Lahav, Omer Nevo
Employees: 40
Funding: $80 million from Sequoia, Swish, Red Point, Assaf Rappaport, Ofir Ehrlich and others
