Irregular, the startup formerly known as Pattern Labs, is staking its claim as the world’s first dedicated frontier AI security lab. Founded in 2021 by Dan Lahav and Eli David, the company recently closed an $80 million funding round co-led by Sequoia Capital and Redpoint Ventures, valuing the firm at $450 million.
CEO and co-founder Dan Lahav sees a seismic shift on the horizon: “Our view is that soon, a lot of economic activity is going to come from human-on-AI interaction and AI-on-AI interaction. And that’s going to break the security stack along multiple points.”
As AI systems get smarter and more powerful, the risks multiply. The breakthroughs driving advances in medicine, science, and technology also open the door to new cyber threats, including vulnerabilities, bypasses, and even fully autonomous cyberattacks.
Developing a research platform to run controlled simulations
Irregular focuses on practical defence, building a research platform that runs controlled simulations on cutting-edge AI models like GPT-5 and Claude 4. The platform tests how such models could be weaponised in cyberattacks and evaluates their resilience to adversarial exploits. Unlike broader governance discussions, Irregular zeroes in on developing tools to counter real-world threats wherever AI is created or deployed.
“This isn’t about governance or lofty theories,” the company says. “It’s about practical tools that stop threats wherever AI is built or deployed.”
The startup’s tech has already caught the attention of top AI labs: OpenAI leverages Irregular’s assessments in system cards for GPT-3, GPT-4 mini, and GPT-5. Irregular also co-authored a white paper with Anthropic on confidential inference systems, while Google DeepMind researchers have cited the company’s platform in studies on AI cyberattack capabilities.
Irregular’s forward-looking plans include expanding its platform’s capabilities to address emerging AI risks and partnering with governments and industry leaders worldwide. Its team blends AI scientists, cybersecurity experts, and national security veterans, many with proven track records in uncovering vulnerabilities in critical infrastructure.
While competitive startups like Redwood Research and Conjecture operate in overlapping AI safety spaces, Irregular’s specialised focus on frontier AI security, combining deep technical research with adversarial testing, sets it apart.
With a team blending AI scientists, cybersecurity experts, and national security veterans, Irregular is positioning itself as a crucial bulwark against the growing threats posed by advanced AI.
