Can GPT-4o be trusted with your private data? 🔐 While it’s a powerful tool, understanding its privacy implications is crucial. 

On May 13, OpenAI launched the GPT-4o AI model to enhance its ChatGPT chatbot. This new version is highly advanced and more humanlike, capable of solving equations, telling bedtime stories, and even recognizing emotions from facial expressions. While OpenAI promotes its commitment to making these tools freely accessible, experts caution that GPT-4o’s enhanced capabilities could lead to increased data collection by OpenAI, raising privacy concerns.

The company has a mixed history in this area. After its launch in 2020, a technical paper revealed that millions of pages from Reddit posts, books, and the wider web, including personal data shared online, were used to develop the generative AI text system. This led to issues with data protection regulators in Italy, resulting in a temporary ban on ChatGPT in the country last year.

Shortly after the launch of GPT-4o, a demo of the macOS ChatGPT desktop app revealed that the chatbot might be able to access a user’s screen. In July, the same app faced criticism again due to a security issue that allowed easy access to chats stored on the computer in plain text. OpenAI promptly released an update to encrypt the chats, but with the heightened scrutiny on the company and GPT-4o, concerns about privacy persist. How private is the latest version of ChatGPT? Is it worse than previous versions? And is there any way to secure it?

OpenAI’s Privacy Policy

OpenAI’s privacy policy reveals extensive data collection, including personal information, usage data, and content you provide when using their services. ChatGPT utilizes this shared data to train its models, unless you disable this feature in the settings or use the enterprise version.

OpenAI states that individual data is “anonymized,” but their approach seems to be “collect everything now and sort it out later,” according to Angus Allan, senior product manager at CreateFuture. The policy explicitly states that all user input is collected and may be used to train models. This includes images and voice data under the catch-all “user content” clause, making it a comprehensive data collection strategy. Although the policy has not significantly changed with GPT-4o, its expanded capabilities mean that the definition of “user content” has widened considerably.

ChatGPT does not access any data on your device beyond what you input into the chat, but it does collect extensive data by default, notes Jules Love, founder at Spark. This includes everything from prompts and responses to email addresses, phone numbers, geolocation data, network activity, and device information.

 

OpenAI says this data trains the AI model and improves its responses, but the terms allow the firm to share your personal information with affiliates, vendors, service providers, and law enforcement. “So it’s hard to know where your data will end up,” says Love.

OpenAI’s privacy policy states that ChatGPT collects information to create an account or communicate with a business, says Bharath Thota, a data scientist and chief solutions officer of analytics practice at management consulting firm Kearney. This includes full names, account credentials, payment card information, and transaction history. Personal information may also be stored if images are uploaded as part of prompts or if users connect with OpenAI’s social media pages like Facebook, LinkedIn, or Instagram.

OpenAI uses consumer data similarly to other big tech and social media companies but does not sell advertising, says Jeff Schwartzentruber, senior machine learning scientist at security firm eSentire. Instead, it provides tools, using user input data to improve services and enhance the value of OpenAI’s intellectual property.

Privacy Controls

Since its launch in 2020 and amidst criticism and privacy scandals, OpenAI has introduced tools and controls to help secure your data. OpenAI states it is “committed to protecting people’s privacy.”

For ChatGPT specifically, OpenAI acknowledges that users might not want their information used to improve its models and provides ways to manage their data. “ChatGPT Free and Plus users can easily control whether they contribute to future model improvements in their settings,” the company states on its website, adding that it does not train on API, ChatGPT Enterprise, and ChatGPT Team customer data by default.

“We provide ChatGPT users with a number of privacy controls, including an easy way to opt out of training our AI models and a temporary chat mode that automatically deletes chats regularly,” OpenAI spokesperson Taya Christianson tells WIRED.

The company says it does not seek out personal information to train its models, nor does it use public information on the internet to build profiles, advertise, target individuals, or sell user data.

OpenAI does not train models on audio clips from voice chats unless you choose to share your audio “to improve voice chats for everyone,” as noted in the Voice Chat FAQ on OpenAI’s website. If you share your audio, it may be used to train models, and transcribed chats may also be used depending on your choices and plan.

 

In recent years, OpenAI has improved transparency around data collection and usage to some extent, offering users clear options to manage their privacy settings, according to Rob Cobley, a commercial partner at law firm Harper James. “Users can access, update, and delete their personal information, giving them control over their data.”

The simplest way to protect your data is to turn off data collection in your personal settings, says Love. Allan advises that “almost everyone” should take a moment to opt out of model training as soon as possible. “This doesn’t remove your content from the platform but prevents it from being used to train future models, which theoretically reduces the risk of data leaks.”

To opt out, navigate to Settings, Data Controls, and disable “Improve the model for everyone.” You can also use a “Temporary Chat” mode by selecting ChatGPT in the top left and toggling on Temporary Chat from the bottom of the list.

However, limiting data collection may impact functionality. “It will not remember previous chats, so responses may be more generic and less nuanced,” notes Love.

From the ChatGPT web interface, users can delete their chat history, customize privacy settings, control shared links, export data, and delete their accounts, says Schwartzentruber. “For additional security, consider enabling multifactor authentication and logging out of all devices.”

Be aware of other privacy concerns when using ChatGPT, such as the potential exposure of sensitive data through Custom GPTs. Managing interaction data carefully can help maintain privacy, but it may also affect your experience, says Schwartzentruber. “Restricting data sharing can reduce personalization and accuracy, as the AI relies on a more limited set of algorithms.”