We agonize over cookie consent banners and lock down our social media privacy settings. Then we paste tax documents, client proposals, and medical questions directly into AI chatbots without a second thought.
That disconnect is becoming harder to ignore. As AI tools move from occasional novelties to daily work infrastructure, the question of what happens to data users send to these models is getting louder - and the answers aren't reassuring.
On ChatGPT's free and Plus plans, conversations train future models unless you manually disable the setting. Google's Gemini logs chats as part of your Google Activity, feeding back into product development. Microsoft's Copilot follows similar patterns. The common thread across major AI platforms: your input becomes training data by default. Opting out requires knowing the option exists and finding where it's buried.
For casual use - brainstorming headlines or asking for recipe ideas - none of this matters much. But that's not how people actually use these tools anymore. Teams paste in proprietary source code. Freelancers upload entire client briefs. Job seekers feed in resumes with home addresses and phone numbers. That kind of data entering a training pipeline carries real risk, not just for you, but potentially for anyone who receives a model-generated response influenced by your input.
The practical response isn't to stop using AI tools. It's basic data hygiene: review your privacy settings on every AI platform you use, use API access instead of the chat interface for sensitive work, and apply the email rule - if you wouldn't send it to a stranger, don't paste it into a chatbot. None of this is complicated. The hard part is remembering to do it when the tools make it so easy not to think about it.