It can be easy and comforting to think that using AI tools involves interacting with an independent machine that knows nothing about you. But between cookies, device identifiers, sign-in and check-out requirements, and the occasional human reviewer, online services seem insatiable for the high demand for your data.
Privacy is a major concern of both consumers and governments about the spread of AI. Across the board, platforms tout their privacy features—even if they're hard to find. (Paid and commercial plans often don't include training on the provided information at all.) But anytime a chatbot “remembers” something, it can still feel intrusive.
In this article, we'll explain how you can tighten your AI privacy settings by deleting your previous chats and conversations and turning off the settings to allow developers to train their systems in ChatGPT, Gemini (formerly Bard), Cloud, Copilot, and Meta AI. Your data. These instructions are for each desktop, browser-based.
Discussion GPT
Still at the forefront of the generative AI movement, OpenAI's ChatGPT has several features to improve privacy and ease concerns about user queries used to train chatbots.
In April, OpenAI announced that ChatGPT can be used without an account. By default, questions shared in the free, accountless version are not saved. However, if a user does not want ChatGPT to be used for training, they still need to change the “temporary chat” settings in the ChatGPT drop-down menu on the left side of the screen.
If you have an account and a ChatGPT Plus subscription, however, how do you disable your requests? GPT-4 gives users the ability to delete all chats under the general settings. Again, to make sure the chats are not being used to train the AI model, look down to “Data controls” and click the arrow to the right of “Refine model for all”.
A separate “Model Update” section will appear, which you can turn off and select “Done”. This removes OpenAI's ability to use your chats to train your ChatGPT.
However, there are still caveats.
“When history is disabled, new conversations will not be used to train and improve our models and will not appear in the history sidebar,” an OpenAI spokesperson told Decrypt.
Claude
“By default, we do not train our models on user-supplied data,” an Anthropocene spokesperson told Decrypt. “Until now, we have not used any customer or user-supplied data to train our genetic models, and we have clearly defined that in the model card for our Claude 3 model family.
“We can use user queries and results to train the cloud. It helps the AI model “recognize patterns and relationships between words.”
Deleting chats archived in the cloud also makes them inaccessible. “I don't keep or have access to previously deleted conversations or conversations,” the Cloud AI agent answers helpfully in the first person. My replies are based on the current conversation only.
Like ChatGPT, Cloud retains certain information as required by law.
“We also retain information in our backend systems for the amount of time specified in our Privacy Policy unless we need to enforce our Acceptable Use Policy, Terms of Service or policy violations, or as required by law,” explains Anthroposic.
Regarding cloud data collection on the web, an Anthroposic spokesperson told Decrypt that the AI developer's web crawler respects industry-standard technical marks such as robots.txt, which site owners can use to opt-out of data collection and that Anthroposic will not access it. Bypassing password-protected pages or CAPTCHA controls.
Gemini
By default, Google tells Gemini users that “your conversations will be handled by human reviewers to improve the technologies that power Gemini Apps. Don't submit anything you don't want to be reviewed or used.”
But Gemini AI users can delete their chatbot history and opt out of using their data to train the model in the future.
To do both, navigate to the bottom left of the Gemini home page and find “Activity.”
Once on the Activity screen, users can then turn off “Gemini Apps Activity”.
A Google representative explained to Decrypt what the “Gemini Apps Activity” setting does.
If you turn it off, your future conversations will by default not be used to improve our machine-learning models, a company representative said. “In this case, your conversations will be kept for up to 72 hours so that we can provide the service and provide any feedback. During those 72 hours, unless a user chooses to provide feedback in Gemini Apps, it will not be used to improve Google products, including machine learning technology.”
There is a separate setting to clear your YouTube history associated with Google.
Pilot
In September, Microsoft added the Copilot generator AI model to the Microsoft 365 suite of business tools, the Microsoft Edge browser and the Bing search engine. Microsoft has included a chatbot preview in Windows 11. In December, Copilot was added to the Android and Apple app stores.
Microsoft doesn't offer an opt-out option to not have user data to train its AI models, but like Google Gemini, Copilot users can delete their history. The process isn't intuitive on Copilot, but previous chats will still appear on the desktop version's home screen even after they've been deleted.
To find the option to delete Copilot history, open your user profile in the upper right corner of your screen (you must be signed in) and select “My Microsoft Account”. Select “Privacy” on the left and scroll to the bottom of the screen to find the copy section.
Because Copilot is integrated with the Bing search engine, the Clear function also clears search history, Microsoft said.
A Microsoft spokesperson told Decrypt that the technology giant uses a variety of techniques, including encryption, de-identification, and storing and retaining user-related information for as long as necessary.
“A portion of the overall user questions in Copilot and Copilot Pro responses will be used to fine-tune the experience,” the spokesperson added. “Microsoft takes steps to identify data before using it to help protect consumers' identities,” it said.
Meta AI
In April, Meta – the parent company of Facebook, Instagram and WhatsApp – introduced Meta AI to users.
“We're releasing the new version of Meta AI so you can ask any question you have about our apps and glasses,” Zuckerberg said in an Instagram video. Our goal is to build world-leading AI and make it accessible to everyone.
Meta AI does not give users the option to not have their input used to train an AI model. Meta provides an option to delete past conversations with the AI agent.
To do this, on a desktop computer, click the Facebook settings tab at the bottom left of your screen, above your Facebook profile picture. Once in the settings, users have the option to opt out of conversations with Meta AI.
Meta explains that deleting conversations here won't delete chats with other people in Messenger, Instagram, or WhatsApp.
A Meta spokeswoman declined to comment on whether or how users can opt-out of their data being used in Meta AI model training, instead pointing Decrypt to the company's September page on Meta settings about its privacy protections and deleting history.
“Publicly shared posts on Instagram and Facebook – including photos and text – are part of the data used to train generative AI models,” the company explains. “We didn't train these models using people's personal posts. We also do not use the content of your private messages with friends and family to train our IS.
But anything you send to Meta AI will be used for model training — and beyond.
“When we interact with businesses that use Meta AI or Creative AI, we use the information people share to improve our products and for other purposes,” Meta added.
Summary
Among the top AI models we covered above, OpenAI's ChatGPT offers the easiest way to delete history and opt out of chatbot queries to train the AI model. Meta's privacy practices seem pretty vague.
Most of these companies offer mobile versions of their powerful apps that provide similar controls. Individual actions may vary – and privacy and history settings may work differently across platforms.
Unfortunately, even cranking all privacy settings to their tightest levels may not be enough to protect your data, Eric Voorhees, founder and CEO of Venice AI, told Decrypt that it's naive to think your data has been deleted.
“Once a company has access to your information, you can't believe it's ever gone,” he said. “People should assume that everything they write to OpenAI goes to them and they will have it forever.”
“The only way to solve this is by using a service where the data doesn't go into a central repository,” Voorhees added — a service like his.
Edited by Ryan Ozawa.
