Sam Altman, OpenAI’s CEO and the public face of ChatGPT, recently gave an unfiltered reason why people might want to stop using cloud-based AI tools like ChatGPT altogether—at least for deeply personal matters. Appearing on Theo Von’s podcast, Altman highlighted a growing concern many users haven’t fully considered: legal exposure. Everything you tell ChatGPT is saved. And while OpenAI promises privacy and encryption, Altman himself pointed out that nothing stops a court from subpoenaing your conversations.
During the conversation, Altman reflected on how some users—especially younger people—are treating ChatGPT like a therapist, confidant, or life coach, offloading their emotional burdens and complex moral dilemmas to the chatbot. But that poses a serious issue: unlike real therapists, lawyers, or doctors, there’s no confidentiality clause protecting what you say to an AI. “We don’t yet have legal privilege for AI,” Altman said plainly, noting that everything from marital struggles to ethical quandaries could potentially be exposed in a legal proceeding.
That comment, first highlighted by PCMag, underscores a key reason why people are increasingly interested in running local LLMs (large language models) directly on their own PCs. Tools like GPT4All and LM Studio let users operate chatbots offline without ever needing to connect to the cloud. These models run on modern GPUs or even NPUs built into newer laptops, and the data they process never leaves the device unless the user decides otherwise.
The appeal is clear: total control and local privacy. Unlike a centralized service, where the company controls your data retention and could be forced to disclose it, a local LLM leaves the responsibility entirely with the user. You can delete sessions, erase logs, or avoid saving anything at all. And while your PC could still be subpoenaed or searched with a warrant, that’s a significantly higher bar than a quiet legal request made to a company server.
Altman’s remarks don’t mean people should stop using ChatGPT altogether. It remains an immensely powerful tool for creativity, problem-solving, and productivity. But if your interactions involve sensitive personal topics—things you’d only tell a human therapist—then a local AI model, or better yet, a real human expert, might be the more appropriate route. After all, privacy isn’t just about encryption—it’s about knowing who controls your information.
