As children increasingly use artificial intelligence (AI) tools, such as ChatGPT, Google Gemini, Claude, Meta AI and Microsoft Copilot, for learning, research and other activities, it is important to recognise that websites, links, files or copied content may contain hidden or misleading instructions designed to manipulate these tools. This is known as a prompt injection attack.
Such instructions may manipulate the AI tool into producing unintended responses, requesting personal information from the child or, where the tool has access to such information, exposing it. This may include names, phone numbers, email addresses, locations, account details or passwords. In some cases, the harmful instructions may be hidden from the child or difficult to recognise.
Parents and guardians should encourage children to:
- Avoid uploading unfamiliar files to AI tools.
- Avoid asking AI tools to open or analyse unfamiliar links.
- Avoid copying content from untrusted websites into AI tools.
- Never enter passwords, account details or other sensitive information into AI tools.
- Seek guidance from a parent, guardian or trusted adult when unsure about online content.
Parents, guardians and caregivers should guide children in using AI safely and remind them that not every file, link or piece of online content is safe to use with an AI tool.