ChatGPT users discovered a bug that allowed them to view the conversations of other users.

Image Credit: Mid-Day A strange bug that allowed users of ChatGPT to view the chat history titles of other users was reported by users. It serves yet another sobering reminder that contrary to popular belief, other people, including the company itself, have access to the messages you send to your neighborhood chatbot. It is just another quirk of the closed environment of OpenAI's wildly successful AI chatbot. A few ChatGPT threads on Reddit and Twitter demonstrated how a sidebar that typically displays user history was also displaying the history titles of other users, on Monday. It is unclear why the Reddit user was coming across some works in Chinese and histories involving Chinese ideologies. A much wider range of prompts was provided by cybersecurity consultant Jordan Wheeler in a Monday Twitter post. Later, he clarified that he was additionally experiencing numerous network connectivity-related errors. Early on Tuesday, an Open AI representative told Bloomberg that the bug has since been fixed. To address the bug, OpenAI temporarily disabled ChatGPT; the chatbot was then restored late on Monday. Even though an investigation is still ongoing, the company claimed it was caused by "unnamed, open-source software.". Image Credit: Beebom Users' ChatGPT history is still "temporarily unavailable" as of early Tuesday reporting time, but the company is "working to restore this feature as soon as possible." The GPT-4 language model from OpenAI, which was just released, powers ChatGPT. It is unclear whether the problem was in any way connected to the change from the old model to the new one. Read More: User Data is Being Stolen by False ChatGPT Apps: Delete These Apps Right Away Users were prevented from viewing other users' chat histories while the bug was active, the company told Bloomberg. It nevertheless highlights a rather alarming aspect of OpenAI's closed-door system. The company can see what users are talking about on chat. Even though the company claims it removes personally identifiable information from the data, regular users are still fair game despite its assurances that it will not use the data from businesses that pay for its API. Users must submit a form with an organization ID if they do not want their information to be used to train the company's chatbot or DALLE-2 art generator. Sam Altman, CEO of OpenAI, told ABC News last week that there is a genuine concern that the LLM could be misused, despite boasting about how excellent their Black Box large language model is. This was in response to the release of GPT-4. In that regard, researchers like those at Check Point Research are already concerned about how the system could be used to generate malicious code.