I think it's a given that governments, PR/brand management firms, and other large actors are racing to weaponize chat generators. The only reason they're likely not already being deployed at scale is because the current stuff isn't really fit for purpose. Being able to say some "appropriate" things most of the time doesn't matter when you spaz out against unsophisticated adversarial actors, even to the point of spitting out your own classified training prompts.

They need to be able to vary their writing style, vernacular, content, subtly (and consistently) reflect an imaginary backstory, following invisible/unspoken standards, and more. All on top of not spazzing out. So we're still somewhat safe from our robot overlords, but they're probably coming soon. And it's only then that we'll get to measure their impact.