It's fearmongering intended to push regulation to ban foreign models (i.e., competition) and locally run ones, and to prevent new AI companies from gaining traction, thus guaranteeing a monopoly for the current subscription based ones.

Also to "place limits on how powerful the models can be", so they have an excuse when they reach the limits of what LLMs are capable of or model collapse starts making them worse, whatever comes first.

In both cases, futile attempts at ensuring their survival once the bubble bursts.

Also, astroturfing. Marketing. Making their models look more powerful. Dangling forbidden fruit in front of their customers and their competitors' in order to entice them, feed their AI psychosis, and strengthen their addiction.