AI has changed what privacy means. Data could always be misused, sold, or stolen; what’s new is the scale and sophistication of inference. A language model trained on your emails can reconstruct your personality, relationships, political views, and health concerns from patterns you’d never consciously notice. A system fed your purchase history, location, and browsing can infer things you’ve told no one.
The answer isn’t technophobia. It’s practical, proportionate measures that cut your exposure without making digital life unusable.
Understand your threat model
Privacy measures aren’t one-size-fits-all. Ask yourself three things. Who are you protecting against: advertisers and data brokers, your employer, a stalker, a government? What data matters most: financial information, location, health data, professional communications? And what are you willing to trade off in convenience, cost, or time?
Most people’s threat model is data brokers selling their information, targeted advertising, and the general risk of a breach. The measures below address that. People in higher-risk situations (journalists, activists, abuse survivors) need more, and should consult resources like the EFF’s Surveillance Self-Defence guide.
Communications
For messaging, Signal is the standard. Messages are encrypted on your device, Signal can’t read them, and it can’t be compelled to produce them. The protocol is open source and audited. Use it for anything sensitive.
Email is more complicated. Standard email (Gmail, Outlook) is processed by the provider and routinely used for ad targeting, spam detection, and, depending on jurisdiction, law enforcement access. For genuinely private email, ProtonMail and Tuta offer end-to-end encryption and zero-knowledge storage at rest, at the cost of limited integration with external clients. SimpleLogin is an aliasing service that creates disposable addresses you forward to your real inbox, then block and delete when they start getting spam.
For most people: use Gmail for low-stakes mail, use aliases for signups so you can trace and cut off spam sources, and use Signal for anything you wouldn’t want a third party to read.
Standard Zoom and Teams calls aren’t end-to-end encrypted by default. For sensitive calls, Signal’s video calling or a self-hosted Jitsi Meet are encrypted alternatives.
Data minimisation
The most effective privacy protection is data that doesn’t exist.
Use email aliases for every signup. SimpleLogin, Apple’s Hide My Email, and DuckDuckGo Email Protection all create a disposable address per service. When a company sells your data or gets breached, the exposure is contained to that alias, which you can delete.
Use a password manager. Unique, strong passwords per service limit breach exposure, since a leaked password from one service can’t open another. 1Password and Bitwarden are both excellent.
Opt out of data broker profiles. Brokers like Spokeo, BeenVerified, and Whitepages aggregate and sell personal information, and in many jurisdictions you can opt out. DeleteMe and Incogni automate removal requests across the major brokers; they aren’t free, but they save a lot of time.
Review app permissions. On iOS and Android, check now and then which apps have location, microphone, camera, and contact access. Most ask for more than they need, so revoke anything that isn’t necessary for the app’s core function.
Browsing
Encrypt your DNS. Your DNS queries (the lookups that turn domain names into IP addresses) are visible to your ISP by default. DNS over HTTPS, using Cloudflare’s 1.1.1.1 or Quad9’s 9.9.9.9, encrypts them, and most modern browsers and operating systems support it natively.
Use a privacy-respecting browser. Firefox with uBlock Origin is the most practical desktop choice, Brave is a reasonable alternative, and Safari has strong tracking protection built in. Chrome’s Privacy Sandbox changes haven’t removed the basic tension between a Google browser and user privacy.
Pick a private search engine. DuckDuckGo, Brave Search, and Kagi are the main options. Kagi (paid) has the best search quality of the three; for general use, DuckDuckGo is good enough and free.
Guard against fingerprinting. Your browser exposes a near-unique fingerprint from its configuration, installed fonts, screen resolution, and dozens of other attributes, which can track you without cookies. Privacy Badger (EFF), Brave’s built-in fingerprint protection, and Firefox’s resistance mode all help.
AI tools and your data
AI tools raise their own privacy concerns.
Whatever you put into an LLM goes somewhere. Pasting a document into ChatGPT, Claude, or Gemini sends that text to a third party, and most providers use it to improve their models unless you opt out or use an API key with a data opt-out. Read the privacy policies. For sensitive work (legal documents, financial data, personal information), use a self-hosted model (Ollama with Llama or Mistral runs on a modern laptop) or an enterprise tier with clear data-handling terms.
Voice assistants do continuous audio processing for wake word detection. Whether that happens on-device or in the cloud depends on the product and the setting. If it worries you, turn voice assistants off when you’re not using them, or use a model that processes on-device, as recent Pixel and Apple devices do for basic queries.
Many consumer apps (photo editors, note-taking tools, email clients) have added AI features that upload your content to the cloud. Check what’s being sent and whether it’s needed for the feature you’re using.
Device security
Turn on full-disk encryption. It’s on by default on modern iOS and Android devices; on macOS it’s FileVault and on Windows it’s BitLocker. Encrypted storage means a stolen device’s data is inaccessible without your password.
Use strong passcodes. Six-digit PINs are crackable, and alphanumeric passcodes are far stronger. On a phone, the PIN is the only thing between your data and whoever has the device.
Use 2FA everywhere, and prefer an authenticator app (Authy, 1Password’s TOTP, or the Apple and Google built-ins) over SMS, which is vulnerable to SIM-swapping. Hardware keys like a YubiKey are the strongest option for critical accounts.
Keep software updated. The large majority of successful exploits target known, already-patched vulnerabilities, so auto-updates for your OS and apps are your first line of defence.
The pragmatic position
Perfect privacy is incompatible with a useful digital life. The goal isn’t to disappear; it’s to be a less attractive target and to limit the blast radius when something goes wrong.
Start with the high-impact basics: a password manager, unique email aliases, a VPN on public networks, and privacy-respecting defaults in your browser. Each takes under an hour to set up and cuts your exposure a lot. Add more as your threat model calls for it.
The AI systems that process personal data will keep getting more capable. The data you don’t give them can’t be used against you.