OpenAI on the defensive after agent security breaches

af Humans Versus Robots