A new feature in OpenAI’s ChatGPT chatbot has raised eyebrows among cybersecurity experts and everyday users alike. The company is testing a “Writing Style” feature that allows ChatGPT to learn how you write by referencing examples from your connected apps, including email services like Gmail and messaging platforms like Slack.
This capability sounds similar to a personalization feature offered by Anthropic’s Claude AI model, but takes it a step further by letting ChatGPT draw on existing writing samples within connected services. In theory, this could give ChatGPT a much larger set of real examples to work with, allowing it to better mimic your tone and style when generating new content.
The feature is currently available to a small group of users, who are being asked to provide examples from their connected apps across three categories: messaging, documents, and email. Once configured, ChatGPT can supposedly write in your voice by referencing these samples, rather than requiring you to repeatedly explain your preferred tone or paste examples into the conversation.
But what does this mean for security? For one thing, it raises questions about data sharing between connected services. When you link your apps to ChatGPT, are you giving the company access to sensitive information that could be used in targeted attacks or phishing campaigns? And how will OpenAI ensure that the data is not misused by bad actors?
Another concern is the potential for over-personalization, where ChatGPT becomes too reliant on a user’s writing style and loses its ability to generalize knowledge. This could lead to a situation where the AI model is overly biased towards one particular tone or voice, making it less effective in certain contexts.
OpenAI has confirmed that it’s testing the Writing Style feature, but hasn’t provided any information on when it will roll out to users widely. In the meantime, users who are part of the testing group should be cautious about what they share with ChatGPT and ensure that their connected apps have robust security measures in place.
Ultimately, this development highlights the ongoing trade-off between convenience and security in the world of AI-powered chatbots. While features like Writing Style may make our lives easier, we need to carefully consider the risks involved before embracing them.
Source: Bleeping Computer — 2026-09-07