A VPN is great at hiding where you are and what your internet provider sees. However, when you’re chatting with ChatGPT or Claude, your privacy depends on something far more complicated than a tunnel through the internet.Even the best VPNs won’t stop AI companies from reading, storing, and learning from everything you type. In this article, we’ll unpack exactly what data these platforms see, why encryption has limits, and what you can actually do to protect yourself.What chatbots see (and what a VPN actually hides)(Image credit: OpenAI / Future)Without a VPN, OpenAI (ChatGPT) and Anthropic (Claude) can typically see a lot about your connection. Your IP address reveals your approximate geographic location. Device metadata follows alongside that, including your browser type, operating system, and unique device identifiers.According to OpenAI’s privacy policy, the company collects your IP address; device information, including unique identifiers; cookie data; and your location, along with your email address and payment information.Anthropic does the same. Their policy covers location, connection details, and device usage patterns, especially on the mobile app. Different wording, same result: both companies can see roughly where you are and what you’re using to talk to them.A VPN helps with some of this. It successfully masks your real IP address from the AI platform and encrypts your traffic so your ISP can’t see that you’re using AI services. However, this is where the protection stops. Once your encrypted tunnel ends at the VPN server, the AI company still sees everything that arrives at their doorstep.The limits of encryption: What a VPN can’t hide(Image credit: Getty Images)This is the critical part many people miss. A VPN protects your data while it travels across the internet, but that protection ends the moment your prompt arrives on OpenAI or Anthropic’s servers. Your message has to be decrypted for the AI to process it.Once decrypted, your data sits on the AI company’s systems, governed by their policies rather than yours. A VPN does nothing to hide the content of your prompts, any files or images you upload, your account details and conversation history, or your payment information. The encryption protects the connection between you and their servers, not what happens afterward. No matter how strong the VPN tunnel is, everything you type is fully readable on the other end.What AIs remember and how to actually stay privateHere’s where things get tricky. Most consumer chatbots keep your conversations on file indefinitely and may use them to train future models. You have to manually adjust settings and delete chats yourself if you want them gone.For ChatGPT specifically, OpenAI may use your content to improve services unless you opt out. You can navigate to Settings > Data Controls, then toggle off the option to improve the model for everyone. However, even with this disabled, chats may still be stored for safety and legal reasons.Claude works similarly. Anthropic requires all consumer users to choose whether their chats can be used for training, and the toggle is on by default. If enabled, data can be retained for up to five years, compared to 30 days if you decline. You can disable the setting anytime in your account preferences or use Incognito mode to keep conversations out of your saved history. Paid business tiers come with stricter privacy agreements by default.The verdict and privacy-first alternativesA VPN is a solid first step for network security in that it hides your location from AI providers and keeps your ISP from monitoring your activity. However, it won’t stop AI companies from reading, remembering, and learning from your prompts. If privacy matters to you, adjust your settings by turning off data training in both ChatGPT and Claude, use ChatGPT’s temporary chats or Claude’s Incognito mode, and avoid sharing sensitive information.For genuinely private AI interactions, consider Proton AG’s Lumo. Built by the team behind Proton Mail and Proton VPN, it uses zero-access encryption so only you can decrypt saved conversations, keeps no logs on their servers, and doesn’t train on your data unless you opt in. It runs on European infrastructure under GDPR and Swiss privacy laws.One step further than this is to run open-source models offline with tools like Ollama or LM Studio, keeping everything local so nothing ever leaves your machine.

The September Apple event is the focal point of the consumer tech world, and today is the big day for Apple fans and tech enthusiasts alike to see how the House that Jobs Built will shape the next year of personal technology.We're expecting the launch of the new iPhone 17 family, updated AirPods Pro 3 earbuds, and the latest Apple Watch Series 11 and Apple Watch Ultra 3 wearables. But this won't just be a day for sequels, we're expecting an all-new concept from Apple today with the iPhone 17 Air.Our News Chief Jacob Krol and our Editor-at-Large Lance Ulanoff are in Cupertino at the Steve Jobs Theater in Apple Park report live from today's showcase. The wider TechRadar team in New York, London, and across the globe are following the news with sharp insights and experience commentary.Without further adieu, let's get the Apple party started...Jump to the live updatesWhat you need to knowWe predict the iPhone 17, iPhone 17 Pro, and iPhone 17 Pro Max will launch todayAn iPhone 17 Air has been heavily rumored to replace the Plus modelWe expect a new look, new Apple processors inside, new camera capabilities, and much moreThere should also be an Apple Watch 11 with a tweaked design and improved health monitoring, plus a load of AI featuresThe new Apple Watch Ultra 3 could arrive with satellite connectivity, high-blood-pressure detection, and a refined designThe long-awaited AirPods Pro 3 update could bring improved audio and health-tracking, perhaps with IR cameras for motion controlsLive updates(Image credit: Future)Hi there, it's Phil Berne your US Mobiles editor reporting from New York City while our News Chief Jake Krol and our eminent Editor-at-Large Lance Ulanoff are on the scene at Apple Park! We'll have all the news soon -- the show starts in less than 15 minutes!

A 120-meter ship could soon host thousands of AI GPUs with direct seawater coolingThe project depends on reused ships to cut both building costs and environmental damageMOL and Kinetics promise flexibility, mobility, and power abundance through powerships and offshore renewablesThe idea of putting an AI-powered facility on a ship used to sound like science fiction - and not long ago, there was even a failed attempt by Del Complex to build a floating "AI nation" that would run itself using artificial intelligence.Now, shipping heavyweight Mitsui O.S.K. Lines (MOL) and Kinetics, the energy transition unit of Karpowership, are aiming to realize something far more grounded.The companies are working together to build a mobile floating data center that could house thousands of AI GPUs while addressing digital infrastructure bottlenecks.MOL and Kinetics outline plans for a floating AI data centerThe two firms recently signed a Memorandum of Understanding to develop what they describe as “the world’s first integrated floating data center platform.”The structure will be hosted aboard a retrofitted vessel, supported by a power supply that includes power ships, floating power plants developed by Karpowership, as well as other sources like solar farms, offshore wind, and onshore grids.“This project represents a major step toward our vision at Kinetics, delivering innovative, efficient, and sustainable infrastructure solutions that meet the energy needs of today and tomorrow,” said Mehmet Katmer, CEO of Kinetics.“By pairing mobile power generation with floating data infrastructure, we are addressing critical market bottlenecks while enabling faster, cleaner, and more flexible digital capacity expansion.”The data center is projected to offer between 20 and 73MW of capacity, cooled by direct water systems drawing from seas or rivers.It would be mounted on a 120-meter-long ship, with network plans that include submarine cables and land-based internet exchanges.“This MOU represents an important step forward in using the MOL Group's assets and extensive expertise in ship operations to rapidly build digital infrastructure while minimizing environmental impact,” said Tomoaki Ichida, Managing Executive Officer of MOL."Moving forward, we will continue to expand a diverse range of social infrastructure businesses centered on the shipping industry.”A mobile, sea-cooled, power-rich platform that bypasses land constraints and permitting headaches offers an attractive alternative to overburdened terrestrial data centers.The flexibility is notable, but the scale of the ambition raises questions, and this warrants skepticism.Although the idea sounds perfect on paper, its real-world execution could face the same issues that has plagued similar utopian infrastructure concepts.The MOU promises operations by 2027, contingent on “successful feasibility studies and ongoing technical developments.”Those feasibility studies will need to prove that issues like network latency, physical relocation risks, regulatory uncertainty, and long-term maintenance can be reliably addressed.Cost and sustainability arguments hinge heavily on the reuse of existing ships.“In addition to reducing construction costs,” the project claims, “the use of existing onboard systems... is expected to reduce initial investment costs.”By avoiding new builds, the developers believe they can also cut the environmental toll of raw material extraction.The practical advantages, such as speed of deployment, mobility, and independence from strained land-based grids, are not in doubt.“Even in areas experiencing power shortages, offshore data centers can begin operations immediately,” the developers note.But whether this system will prove reliable, scalable, and economically sound in the long term remains to be seen.You might also likeTake a look at our picks for the best WFH app aroundCheck out our recommendations for the best monitors for working from homeMegawatt-class AI server racks may well become the norm before 2030