Irregular Introduces Frontier AI Security, Raises $80 Million

Irregular, the first frontier security lab with the mission of protecting the world in the time of increasingly capable and sophisticated AI systems, led by Dan Lahav and Omer Nevo, has introduced Frontier Security Lab and raised $80 million led by partners at Sequoia and Redpoint.

AI is advancing faster than expected. The same breakthroughs enabling medicine, science, and software also create risks: systems that can discover exploits, evade defenses, or even carry out autonomous cyber operations. These are not theoretical dangers but the natural byproduct of pushing intelligence to its limits.

Irregular was founded to address this reality, building a research platform for controlled simulations on frontier AI models, evaluating both their potential for misuse in cyber operations and their resilience when under attack.

Already shaping the field: OpenAI cites evaluations in system cards for GPT o3, o4-mini, and GPT-5; Google DeepMind cited and used the platform in research on emerging AI cyberattack capabilities; a seminal paper with RAND on securing model weights helped set global AI security practice; collaborations include a white paper with Anthropic on confidential inference; and partnerships extend to governments such as the UK for vetting frontier-model cyber capabilities.

Irregular is creating the defensive systems to secure the next generation of AI, not through governance or theory, but with practical tools that stop threats wherever AI is built or deployed. By combining applied AI research with nation-state cybersecurity expertise, the company is pioneering frontier AI security as a new category and the first lab of its kind. The team brings exceptional depth, from AI scientists published in top venues to leaders who have scaled R&D at scale, world debate champions, and national security experts who have exposed vulnerabilities in critical infrastructure.