Irregular, which helps AI labs test their models' for misuse, raised $80M across seed and Series A rounds led by Sequoia and Redpoint at a $450M valuation
Frontier AI labs are relying on a small Israeli startup to run their AI through simulations to test its capacity for malicious hacking.
Context & Ripple Effects
AI security has already attracted venture backing at different layers: Robust Intelligence emerged to secure model deployment, while Endor Labs scaled scanning for vulnerabilities in AI-generated code. Irregular sits earlier in that chain, assessing whether frontier models can be induced to cause harm before deployment.
The financing matters because it backs a specialized external testing function for frontier labs, rather than another model builder or application provider. That makes misuse evaluation a more distinct part of the AI development stack.
First-order effects
- Irregular gains capital to expand simulation-based assessments of frontier models’ malicious-hacking and other misuse capabilities.
- Frontier AI labs using Irregular have a better-funded external provider for adversarial testing, while Sequoia and Redpoint add exposure to the AI-safety tooling layer.
Second-order effects
- Security vendors serving AI developers may face pressure to distinguish deployment security and code scanning from pre-release model-misuse evaluation; the services can be complementary rather than interchangeable.
- As labs rely on external simulations, demand can shift toward repeatable testing workflows and specialist evaluators, not solely in-house safety teams.
Third-order effects
- If external testing becomes a routine checkpoint for frontier systems, AI safety may develop into a dedicated vendor category alongside model customization and deployment security.
- The pattern supports broader frontier-AI institutionalization: capital is moving into the operational controls around powerful models, though it remains unclear how standardized external evaluations will become.
The trend: Frontier AI is spawning a specialized assurance ecosystem in which independent vendors test, secure, and operationalize models around their core capabilities.