ChatGPT and other AI assistants are moving beyond simple question-and-answer tools and into our messages, files, calendars, and work routines. OpenAI’s new Mac Messages integration can search conversations, draft replies, and with the user’s approval, send them. But there are genuine risks to users that they may not understand.

ChatGPT’s new plug-in for Mac is useful, but it asks for unusually broad access. To let ChatGPT search, summarize, draft, and send Apple Messages on a Mac, users must opt in and grant the plug-in permissions that go well beyond a typical writing assistant. Employers should add this to their AI policies: do not connect it to work devices or sensitive personal accounts, use it only for low-risk tasks, and make settings require approval to send a message. OpenAI warns that persistent approval removes the final chance to review what is sent in your name.

Do not use personal AI accounts for business. They don’t handle data the same way. OpenAI says ChatGPT Team, Enterprise, and API inputs and outputs are not used to train its models by default, while consumer users must adjust data controls or use Temporary Chat if they do not want new conversations used for training. At home, avoid entering passwords, financial details, health information, private messages, or anything you don’t want entered for helping compile answers to others’ questions.

And remember, answers from ChatGPT and other chatbots often contain errors. Whether you use AI at work or at home, verify facts, sources, calculations, and any message before you rely on it.