Fault in ChatGPT prompts AI privacy concerns

According to OpenAI CEO Sam Altman, a glitch on the platform led to some users viewing the headings of other parties’ conversations. This has led to questions over privacy on the platform, with many people worried that their private information could end up getting inadvertently passed on through the tool.

People swapped pictures of chat histories on social media platforms Twitter and Reddit, with those implicated claiming they had nothing to do with the conversations involved. Some of the titles of the conversations that users reported seeing included “Chinese Socialism Development” and dialogues in Mandarin.

The error suggests that OpenAI has free access to user conversations, which has raised concerns about how seriously the company takes privacy. While the firm’s policy does state the possibility of user data being drawn upon to help develop the training model, it also says that this is only done after PII (personally identifiable information) has been deleted.

The glitch was fixed by OpenAI, but many are not fully convinced of the platform’s ability to uphold privacy and security. The company’s chief executive has promised a “technical post-mortem” soon to deep-dive into the error and how a solution has been formulated.

The problem appears as tech giants such as Google and Microsoft race to launch their own artificial intelligence tools on the market. Google recently gave beta testers and journalists a taste of its new chatbot, Bard, while Microsoft is pumping more money into OpenAI.

The speed of new AI-related features and tools has left critics fearing of how mistakes could have far-reaching negative impacts.

Despite the error, millions of people have used ChatGPT since it launched in November of last year to design messages, write recipes and even pen poetry and short stories. Each interaction with the chatbot is held an individual’s chat history columns for use at a later time.  

As the market for AI expands, it is vital that companies to prioritise privacy and security to prevent similar errors from happening in the future.