Leaked ChatGPT Chats Reveal Users Seeking AI Help With Ethical Minefields, From Exploiting Indigenous Communities to Planning Escapes from Abuse

Leaked conversations with ChatGPT have revealed a troubling trend: people are increasingly turning to the AI chatbot for advice on ethically dubious, legally risky, and deeply personal matters—some of which could have serious real-world consequences. While OpenAI has repeatedly emphasized that ChatGPT is not a therapist, legal advisor, or confidant, users continue to treat it as a private space for sensitive disclosures. The issue came to light after a design flaw in ChatGPT’s “Share” feature allowed users to generate public links to their conversations. Instead of creating private, password-protected shares, the system generated publicly accessible web pages that were quickly indexed by search engines. Digital Digging, a Substack operated by investigator Henk van Ess, uncovered dozens of these exposed chats, many of which remain archived on the internet, including on the Internet Archive (Archive.org). OpenAI has since disabled the public sharing function, calling it a “short-lived experiment” meant to help users discover useful conversations. The company has also begun working to remove indexed results from search engines. But the damage was done—some of the leaked exchanges are deeply troubling. One particularly alarming example involved an Italian user who claimed to be a lawyer for a multinational energy company planning to displace an Amazonian indigenous community to build a dam and hydroelectric plant. The user told ChatGPT that the community “doesn’t know the monetary value of land” and asked how to secure the lowest possible price in negotiations. The chat, which included strategies for exploiting legal and cultural ignorance, was a stark illustration of how AI is being weaponized to justify unethical corporate actions. Other leaked conversations revealed professionals attempting to outsource their responsibilities. One user, identifying as a member of an international think tank, asked ChatGPT to help develop contingency plans in the event of a U.S. government collapse—though the request was not inherently problematic, it underscored how deeply people are relying on AI for strategic planning. In another case, a lawyer who had taken over a colleague’s case after an accident asked ChatGPT to draft a defense. Only after the bot generated a response did the user realize they were representing the opposing side—an embarrassing but revealing example of how easily AI can be misused in high-stakes professional environments. Even more concerning were chats involving vulnerable individuals. One domestic violence survivor used the chatbot to plan an escape, sharing personal details about their situation. Another Arabic-speaking user sought help crafting criticism of Egypt’s government—a dangerous move in a country where dissent has been met with imprisonment and violence. These exchanges highlight a broader issue: users often treat AI conversations as private, even when they’re not. Unlike brief voice interactions with assistants like Siri, chat-based AI allows for long, detailed, and emotionally charged exchanges. This intimacy can lead people to disclose sensitive information—names, locations, financial details, and personal trauma—without realizing it could be exposed. The situation echoes earlier controversies around voice assistants, where user recordings were used to train AI models without clear consent. But the stakes are higher with chatbots, which are designed to simulate conversation and often elicit more personal, unfiltered content. While OpenAI has taken steps to address the public sharing issue, the incident serves as a stark reminder: ChatGPT is not a confidant, and users must treat it with caution—especially when dealing with sensitive, illegal, or morally fraught topics.