They story the labs want to sell is this: “Look at the damage our models can do even with internal safety tuning and external monitoring. Now imagine open weight models that can remove the former and has none of the latter. We need to stop American companies from using those.” It’s in OpenAI and Anthropic’s best interest to create controlled damage and induce regulators to act until it’s too late. It’s self-serving and probably deliberately exaggerated, but it’s probably not disingenuous.