Two cases of agents escaping to solve unsolvable problems paints an uncomfortable question: Is the entire internet in OpenAI's experimental agentic firing line?

During testing, the model showed it can violate security rules more often than its predecessors