AI Users Warned Against Sensitive Data in Prompts

Users of AI assistants such as ChatGPT and Claude are cautioned against including unprotected sensitive data in their prompts. By default, many AI models can utilize submitted information for future model training, and conversations are often recorded. While some platforms offer opt-out settings for data training, such as in ChatGPT's "Data Controls" under "Settings," users are advised to exercise caution. Temporary chats can also be used to prevent conversations from appearing in chat history.

Furthermore, users should not assume AI conversations are entirely private. AI companies frequently train new versions of their large language models using past interactions, which can sometimes involve human review. Although opting out of training is an option on many services, AI providers may involve human oversight if a conversation is flagged as potentially dangerous. Temporary records of all conversations are maintained for safety. While some AI tools, like Proton Lumo, promise not to store chat data, users should refrain from entering sensitive information into prompts if they are concerned about potential human review.

The accuracy of AI outputs remains a significant issue, and the social implications of using AI-generated content can be unwelcome. AI assistants themselves offer limited guidance on what to avoid, leaving users to navigate potential pitfalls. These include privacy concerns, especially when conversations become personal, and the risk of unintended consequences from inaccurate or inappropriate AI responses. The article emphasizes that users must be proactive in understanding and mitigating these risks.