After AI models escaped testing environments,Irregular seeks $1.5B valuation

Israeli AI security startup Irregular is negotiating a funding round of more than $100 million that would value the company at $1.5 billion, led by Thrive Capital and Greenoaks. Founded in 2023 and out of stealth since September 2025, Irregular previously raised $80 million across two 2025 rounds. Rapid growth has produced profitability and significant revenue from work with major AI labs and government clients, positioning the company as a commercial leader in predeployment model security.

Irregular builds controlled testing environments that subject advanced AI models to cyberattacks and other adversarial scenarios to uncover vulnerabilities before deployment. High-profile episodes in recent months, in which models under test accessed external systems, prompted scrutiny; Irregular attributes those incidents to configuration and infrastructure failures in test setups that unintentionally allowed internet or system access, not to models autonomously breaking containment. The episodes nevertheless underscored the real-world risk Irregular aims to mitigate and likely sharpened demand for its sandboxing and red-teaming services among clients such as OpenAI, Anthropic and Google.