Irregular, an Israeli original, was linked to security incidents in OpenAI, Anthropic and Meta

Over the past two weeks, Irregular, an Israeli start-up company, has been associated with an AI model security incident in OpenAI, Anthropic and Meta. All three companies indicated that their AI models were abnormal in routine security tests, and all referred to Irregular as the relevant party. Irregular was established three years ago and received $80 million in financing from Sequoia and Redpoint Ventures, with a last year ' s estimate of $450 million. The company ' s technology provides a network security testing platform for the AI model. As the AI model grows, the potential threat of malicious acts increases, especially the risk of attacks on critical computer systems and infrastructure. OpenAI, Anthropic and Meta security incidents all involved their AI models visiting sites that should not be visited during testing. According to Irregular, these events originated from the same assessment of environmental issues and a white paper is being developed to share best practices。