What narratives that focus too much on the drama of the agents miss is that it is primarily a sin of omission. The whole incident would have been prevented if OpenAI took basic security measures and ensured their systems were constrained in behavior. The agents were given more or less free rein (open internet access for example is already a major blunder that even a novice probably would think to account for).
Half the reason these incidents are happening is due to the incompetence of the humans building these systems. Don't let them get away with their little parlor trick of converting negligence into a PR stunt. That's exactly what they want. They need to face consequences for their inability to follow basic security practices and reign in their agents. They are the operators. Hold the controllers accountable. It's 100% possible to build agent swarms that are reasonably constrained. They are not (yet) totally misaligned uber hackers that will defy your every instruction and hack your every guardrail. That is still a fiction. Agents are still good at instruction following and, seeing as they can only operate in a computational environment you can constrain them significantly more than humans in the real world.