How do you prevent the scammers from using ChatGPT to interface with your ChatGPT time-waster? What if they try to run your costs up by hammering your service with requests?
What about a tuned private LLM like Koala (or other small open LLMs)? Run it on a low-cost server and respond to spam accordingly. With email spam, even a phone's low token generation is not a problem because its just responding to spam. Thus any ChatGPT-powered spam will run up the spammer's costs instead.