Irregular
Founded in 2023, Irregular is an AI security lab running offensive cyber evaluations and defence development with frontier AI labs, with headquarters in Israel and operations in the San Francisco Bay Area.
Irregular is an AI security laboratory founded in 2023 and described as the first of its kind at the frontier of the field. Its work centres on the threats posed by increasingly capable AI systems, combining offensive cyber evaluations with defence development. The company runs this work in partnership with leading AI laboratories, including Anthropic, OpenAI and Google DeepMind, and operates from headquarters in Israel with activities in the San Francisco Bay Area.
Its technical domains cover AI security, cybersecurity evaluations, red teaming and attack simulation, AI model risk assessment, and safety and responsible deployment. Among its products is the SOLVE framework, a method for vetting risks in AI models that has been adopted by the UK government and by Anthropic, including to vet risks in Claude 4. The company also maintains a proprietary set of cybersecurity evaluations that place AI models in realistic attack scenarios; these have been cited in system cards for GPT-4, o3, o4 mini, and 5. A further offering, the Scenario Evaluation Suite, simulates complex exploitation chains within attack scenarios modelled on real-world operations.
Irregular has raised $80 million at a valuation of $450 million. The company describes an embedded, side-by-side working model with frontier AI research labs and a mission-driven focus on defending against next-generation AI threats. It positions itself within the discussions shaping responsible AI deployment and the setting of security standards for the AI industry.