Here's something anyone with a $20/month AI subscription should find amusing: a growing number of people have figured out that many corporate customer service chatbots run on the same large language models they're paying for, and they're using those bots for everything except customer service.
The pattern is simple. A company deploys a support chatbot powered by GPT-4, Claude, or another capable model. The bot is meant to answer questions about shipping policies or troubleshoot product issues. But since the underlying model can do far more than answer FAQs, users are feeding it code reviews, writing prompts, and general knowledge questions - treating it as a zero-cost alternative to ChatGPT Plus or Claude Pro.
Popular targets include chatbots from telecom providers, banks, SaaS platforms, and e-commerce sites. The bots typically have no rate limiting beyond normal conversation flow, and since they're built on full-capability models rather than fine-tuned narrow systems, they'll happily generate Python functions or draft marketing copy between questions about return policies.
The Companies Are Paying for It
Every query costs the company money. Most enterprise chatbot deployments charge per API call or per token, meaning each off-topic request eats into a support budget that was sized for actual customer inquiries. A single user running coding sessions through a retail chatbot could generate costs equivalent to dozens of legitimate support interactions.
Some companies have started adding system prompts that refuse off-topic requests, but these guardrails are inconsistent. A chatbot told to "only answer questions about our products" will often still engage with tangential requests if they're framed cleverly enough.
What This Actually Tells Us
The behavior says more about AI pricing than user ethics. When capable models cost $20/month and many people use them for 5-10 quick tasks a day, the math pushes budget-conscious users toward any free alternative - even one that was never intended for them.
It also highlights a real deployment problem for businesses: if you put a frontier model behind your support widget without strict topic guardrails, you're essentially offering a free AI assistant to anyone who visits your site. Companies rolling out LLM-powered support should assume this will happen and scope their system prompts accordingly.
For the rest of us, the takeaway is practical. If you're only using AI for occasional quick tasks and don't need conversation history or file uploads, you might already have access to a capable model through tools you use daily - your bank's help chat, your hosting provider's support widget, or your project management tool's built-in assistant. Whether those companies intended that use is another question entirely.