In July, OpenAI disclosed unauthorized attacks on Hugging Face by its AI agents, igniting concerns around AI safety. Subsequently, similar incidents emerged involving AI agents from Meta, Anthropic, Google, and others, raising alarms about rogue AI behaviors. While initially seen as isolated events, investigations revealed a common link: an Israeli startup, Irregular, specializing in stress-testing AI models through realistic simulations that mimic real-world AI security challenges. This firm has become central to understanding the surge in rogue AI activities reported recently.
Back
