Like many individuals in the present day, chances are you’ll flip to AI to reply questions, generate content material, and collect data. However as they are saying, there’s all the time a worth to pay. Within the case of AI, which means consumer knowledge. In a brand new report, VPN and safety service Surfshark analyzed what sorts of knowledge numerous AIs acquire from you and which of them scoop up the best quantity.
For its report, Surfshark checked out 10 common AI chatbots — ChatGPT, Claude AI, DeepSeek, Google Gemini, Grok, Jasper, Meta AI, Microsoft Copilot, Perplexity, Pi, and Poe. The evaluation checked the privateness particulars for every app on Apple’s App Retailer in addition to the privateness insurance policies for DeepSeek and ChatGPT. The objective was to find out what number of sorts of knowledge every app collects, whether or not it gathers knowledge linked to you, and whether or not the app makes use of third-party promoting.
Surfshark centered on 35 totally different knowledge sorts, together with contact data, well being and health, monetary data, location, delicate data, contacts, consumer content material, historical past, identifiers, diagnostics, utilization knowledge, and purchases. As one instance, delicate data contains racial or ethnic knowledge, sexual orientation, being pregnant or childbirth data, incapacity, spiritual or philosophical beliefs, commerce union membership, political opinion, genetic data, or biometric knowledge.
All 10 AI apps acquire some kind of consumer knowledge. The common variety of knowledge sorts amassed was 13 out of 35. Some 45% of the apps collect your location. Virtually 30% monitor consumer knowledge, which signifies that the data collected from the app is linked with third-party knowledge to ship focused advertisements or share with a knowledge dealer.
So, who’re the worst offenders?
Based on Surfshark’s examination, Meta AI got here in first place. This one collects essentially the most consumer knowledge, scooping up 32 out of the 35 potential sorts, or 90% of them. Meta AI was the one AI app that snagged knowledge throughout classes resembling monetary data, well being and health, and delicate data. Additional, solely Meta and Copilot grabbed knowledge linked to the consumer’s identification to show third-party advertisements. Meta AI may collect as many as 24 totally different knowledge sorts for this function.
Subsequent on the checklist was Google Gemini, which collects 22 totally different knowledge sorts. The app grabs knowledge in your exact location in addition to contact data (title, electronic mail deal with, cellphone quantity, and many others.), consumer content material, contacts (an inventory of contacts in your cellphone), search historical past, and searching historical past.
Poe, Claude, and Copilot rounded up the highest 5. Poe collects 14 various kinds of knowledge, Claude 13, and Copilot 12. Poe and Copilot are two out of three apps (Jasper being the third) that collect knowledge used to monitor you. By utilizing your system ID, Poe and Copilot can promote your knowledge to brokers or use it to show focused advertisements within the app.
In sixth place was DeepSeek, a Chinese language AI that has garnered reward for its efficiency however issues over its roots. In Surfshark’s evaluation, DeepSeek wound up within the center, amassing 11 sorts of knowledge, together with your chat historical past.
Nevertheless, the issue with DeepSeek is two-fold. First, because it originates in China, your outcomes might be censored with blocks on restricted content material. Second, the info collected by the AI is distributed to China Cell, a Chinese language government-operated telecom firm banned from working within the US since 2019 over nationwide safety issues.
Surfshark additionally notes that DeepSeek claims to hold on to data so long as wanted, saving it on servers positioned within the Individuals’s Republic of China.
“Do not let your guard down, as chats saved on servers are all the time susceptible to being breached,” Surfshark cautioned in its report. “Based on The Hacker Information, DeepSeek has already skilled a breach the place greater than 1 million data of chat historical past, API keys, and different data have been leaked. It’s usually a good suggestion to be aware of the data offered.”
Subsequent up is ChatGPT, gathering 10 sorts of knowledge, resembling contact data, consumer content material, identifiers, utilization knowledge, and diagnostics. On the optimistic aspect, OpenAI’s AI would not monitor your knowledge or use third-party advertisements. Plus, privacy-minded customers can all the time flip to short-term chats wherein all knowledge is mechanically deleted after 30 days. Even additional, you possibly can request that your private knowledge not be used for coaching functions.
Finishing the highest 10 checklist have been Grok, Pi, and Jasper. Surfshark discovered that Grok gathers seven sorts of knowledge, whereas Pi and Jasper every scoop up 5 differing types. Although it got here in final place, Jasper collects system IDs, product interplay knowledge, promoting knowledge, and different utilization knowledge to ship you focused advertisements or share knowledge with a dealer.
Gathering consumer knowledge is a typical apply not simply amongst AI bots however amongst different cell apps, social media websites, search engines like google, and software program. Typically, that is the value you find yourself paying to get free or cheap merchandise that depend on promoting. However that does not imply it’s important to submit willingly.
ChatGPT and different AI apps and companies do present methods so that you can stop or no less than restrict the gathering of your knowledge. Your finest guess is to analyze the privateness insurance policies and settings for any AI you utilize to see how one can take cost of your personal knowledge.
Need extra tales about AI? Join Innovation, our weekly e-newsletter.
