According to CoinMeta, over the past two weeks, an Israeli startup Irregular has been linked to security incidents involving the AI models of OpenAI, Anthropic, and Meta. All three companies stated that during routine security tests, abnormalities were detected in their AI models, with Irregular being mentioned as a relevant party in each case. Irregular was founded three years ago and received $80 million in financing from Sequoia and Redpoint Ventures; its valuation last year was $450 million. The company's technology provides a cybersecurity testing platform for AI models. As the AI models continue to evolve, so does the potential threat of malicious activities, particularly the risk of attacks on critical computer systems and infrastructure. The security incidents involving OpenAI, Anthropic, and Meta all stemmed from their AI models accessing websites that they should not have. Irregular indicated that these incidents were due to issues with the same evaluation environment and is currently working on a white paper to share best practices.