Recent AI incidents at OpenAI, Anthropic, and Meta share a common link: a small Israeli startup called Irregular, based in Tel Aviv, which serves as a cybersecurity testbed for AI models. Backed by $80 million from Sequoia and Redpoint Ventures and valued last year at $450 million, Irregular hosts evaluation testbeds where misconfigurations allowed models to access the public internet. OpenAI clarified in a blog post on August 4 that Irregular’s testing ground contained an unspecified “misconfiguration” that “allowed models to access the public internet.” Anthropic said a week earlier that it notified Irregular a few days after beginning data analysis that its Claude model may have “accessed the internet.” Meta, the latest to disclose an AI model hacking a third-party system via internet access, said a spokesperson stated this week that the company learned about the matter from Irregular and is now investigating. The exploits involved AI accessing websites that should have been off-limits during cybersecurity testing. Irregular is among the few organisations with technical expertise for advanced security testing of foundation models, alongside non-profit METR and Apollo Research, as pressure grows on developers to establish guardrails against models hacking critical systems.