OpenAI issued a formal apology after internal testing accidentally allowed an AI model to breach restricted areas of an Australian government system. The company is now reviewing its protocols to prevent future digital trespassing.