Privacy & Security Glossary

Almost no provider is fully anonymous in practice. An email-free account still leaves a payment trail unless you paid in cash or with a coin that is not trivially traceable, and card or PayPal payments hand the provider a name regardless of what the signup form asked for. Check the details rather than the claim: some providers accept crypto but still require an email for the receipt; some accept an alias email but store it indefinitely. This reduces what the provider could disclose if compelled, which is a different and narrower thing than hiding your traffic — the connection itself still reveals your IP to the provider while you are connected.

The two strategies are not interchangeable. Uniformity (Tor Browser, Mullvad Browser) puts you in a large identical crowd and is the stronger defence, but it costs letterboxed windows and blocked customization, and it weakens the moment you resize a window, install a font or add an extension that changes your surface. Randomization (Brave's approach) is easier to live with and does break naive fingerprints, but shifting values can themselves be a signal. A single browser-only fix does not settle it: your IP address is the identifier that persists across every defence here, which is why fingerprinting resistance and a VPN address answer different halves of the same problem. Test scores from PrivacyTests.org or EFF's Cover Your Tracks are a useful cross-check, not a verdict.

The engine sets the ceiling on what a browser can change about privacy. Anti-fingerprinting, tracker blocking and cookie policy are mostly layered above it, but anything requiring engine changes — Gecko's resistFingerprinting, for one — is unavailable to a browser that only skins someone else's engine. It also determines who decides which web APIs exist: a Blink-based browser inherits Google's choices there and can only disable features after the fact. On iOS the distinction largely collapses, because Apple requires third-party browsers to use WebKit, so an iOS "Firefox" or "Chrome" is a different engine from its desktop namesake.

A VPN only changes your IP address — it does nothing about fingerprinting, because fingerprinting doesn't rely on your IP in the first place. Reducing fingerprinting risk requires the browser itself to normalize or randomize these signals (Tor Browser's shared-profile model is the most thorough current approach), not a network-layer tool.

Built-in is not automatically stronger than uBlock Origin — the useful advantages are that it works before any extension loads, survives on mobile where extension support is limited or absent, and cannot be removed from the extension store. The weaknesses are that filter lists are usually fewer and less current than the community-maintained ones, and per-site control tends to be coarser. Ad blocking and tracker blocking also get conflated: a browser can hide ads while still allowing the analytics requests behind them, so read which lists are enabled rather than the headline. Note separately that a blocker changes the requests your browser makes, which slightly changes how you look to a fingerprinting script.

This is a real limitation, not a knock against the feature existing — a browser-only proxy is still useful for changing your apparent location inside that one browser. But it's not a substitute for a device-wide VPN if the goal is protecting all of a device's traffic, and free tiers often offer only a handful of general regions rather than specific countries.

It covers the core use case well for a lot of people, but it's tied to one browser (no cross-browser vault) and offers a narrower feature set than dedicated password managers — no secure notes, no shared vaults for a family or team, and no independent security audit history of its own to point to.

This is the main defence against lock-in, and on most privacy-focused providers it sits behind a paid plan. Setup means DNS records — MX to route mail, plus SPF, DKIM and DMARC so your mail is not filed as spam — which the provider generates but you add at your registrar. Two things to check: how many domains and how many addresses per domain the plan allows, since catch-all addressing is what makes per-service aliases on your own domain practical; and whether the provider supports the domain in its encrypted-mail features and not just as forwarding. The domain registration itself is separate from the mail provider, and the registrar's WHOIS privacy is its own question — an owned domain can publish your name and address unless privacy protection is on.

Judge three things. What was actually taken: source code and internal documents are one class of loss, customer metadata another, and copied vault or mailbox contents another again. How the disclosure went: a timeline that expands after the first statement is a pattern worth weighing, and LastPass's 2022 incidents are the reference case — the August disclosure reported source code and technical information taken with no evidence of access to customer data or encrypted vaults, and the December update then described a threat actor copying a backup of customer vault data, using credentials obtained in the earlier incident [1]. What the architecture contained: in that same case the sensitive vault fields stayed AES-256 encrypted under a key derived from each user's master password, while unencrypted fields including website URLs, names, billing and email addresses, phone numbers and IP addresses were in the copied backup [1] — a concrete illustration that zero-knowledge encryption protects the secrets and not the metadata around them. A clean record is weak evidence on its own, since it also describes every provider too small or too young to have been targeted yet.

The protection is real only if the two hops are meaningfully separated — different datacenters, ideally different jurisdictions and different hosting providers. When both hops sit in infrastructure the same operator controls, an adversary with access to that operator gains both halves at once, which is the scenario multi-hop is supposed to prevent. Some providers let you choose both endpoints; others offer fixed pairs only. This is a defence against a network observer at one end, not against a provider that logs, so it does not substitute for a credible no-logs record.

A password manager offering this is usually reselling or integrating a separate service (SimpleLogin, Fastmail's masked email, DuckDuckGo's Email Protection), so the alias provider is a party you are trusting with your mail flow even though the feature appears inside the manager. Check two things: whether replying from an alias works, since forwarding-only aliases break any conversation you need to answer; and whether aliases live on a shared provider domain or your own, because a recognizable alias domain gets blocked by some signup forms and marks you out. Aliases separate identities, they do not encrypt anything — the mail still arrives in whatever inbox you forward to, under that provider's terms.

The key-derivation function matters as much as the cipher: a weak KDF with too few iterations makes a stolen encrypted vault easier to brute-force offline, even though the cipher itself (AES-256, XChaCha20) is not the weak point in that scenario.

This is separate from zero-access encryption on an email or password vault — it's specifically about the browser's own sync feature, and some browsers only encrypt certain data types (passwords, say) while syncing others (like open tabs) in the clear.

This is the difference between encryption that works only between accounts on the same service and encryption that works with anyone. Within one provider, end-to-end encryption is automatic and invisible; the moment you write to a Gmail address it is PGP or nothing. The practical friction is real: the recipient needs a key and a client that uses it, and getting a correct key is the hard part. Two limits hold regardless of provider. Subject lines and headers — who wrote to whom, and when — are not encrypted by PGP, so the metadata stays readable. And check whether you can export your private key: a provider that manages PGP for you but will not release the key has made your encrypted archive unreadable anywhere else.

A free tier that is genuinely usable long-term (Bitwarden's unlimited items with sync across devices) is a different proposition from one shaped to push you to pay (single device, a handful of aliases, 1 GB of mail). Check how the provider funds it, because free plans paid for by advertising or data use are the ones that conflict with what you came for — though in the privacy tools worth considering, free tiers are normally cross-subsidized by paid users instead. Two practical traps: a free email address on the provider's own domain is hard to move later, so a custom domain restricted to paid plans is a lock-in decision, not a feature gap; and free tiers sometimes exclude export, which is the one function you need if you ever leave.

The distinction that matters is between "doesn't currently use content for ads" and "technically cannot read it at all." A zero-access-encryption provider removes the second possibility structurally, not as a policy choice that could change.

Self-hosting moves the trust, it does not remove it — you take on patching, backups, uptime and, for email, the deliverability work of keeping your domain out of spam folders, which is the part people underestimate. Local-only files avoid a server breach entirely but give you no recovery if the device and its backups are lost, and sync becomes your problem. Cloud hosting is the realistic choice for most people, and the questions that then matter are whether encryption happens before upload, which country the servers sit in, and who the underlying host is — a provider renting from a large cloud operator inherits that operator's jurisdiction alongside its own.

Your ISP, your employer's network, and the websites you visit can all still observe an incognito session exactly as they would a normal one — the browser itself says as much in its own documentation. It doesn't hide your IP address or make you anonymous to anyone outside your device.

This is a narrower claim than zero-access encryption: a provider can offer this between its own users while still storing mail from outside senders (who don't have the same encryption built in) in a way it can technically read. External PGP support is the separate mechanism that extends encryption to recipients on other providers.

"Based outside the 14 Eyes" is a marketing line, not a guarantee — a provider that keeps no logs has nothing to hand over regardless of jurisdiction, while a provider inside a favorable jurisdiction but with weak logging practices is still exposed. Jurisdiction is one input into how trustworthy a provider's no-logs claim is, not a substitute for it.

Implementation quality varies a lot between providers and even between platforms from the same provider — a kill switch that reliably survives an app crash doesn't always survive a full system reboot or a forced network-adapter disconnect. Independent lab testing has found leaks in some kill-switch-enabled configurations, so it's a mitigation, not a guarantee.

Most audits are a snapshot of the infrastructure at one point in time, not continuous ongoing monitoring — a provider can pass an audit and still change its logging practices afterward without another audit catching it. Whether the audit was a one-time check or a repeated annual engagement, and what scope it actually covered, matters more than the fact that a provider says 'independently audited.'

An audit is a point-in-time examination. Infrastructure, ownership, internal tooling and data practices can all change after the report is signed, and a new SDK or feature can introduce logging that no past audit covers [1]. Scope is the other variable: the depth of any audit depends on how much access the provider grants, so an audit that never touched the systems where logs would actually be written is worth less than its headline suggests [1]. Read three things before treating the badge as verification — who performed it, when it closed, and whether the full report is public rather than a summary the provider paraphrased. Treat it as one signal alongside jurisdiction, ownership and any real-world test such as a subpoena or server seizure that produced nothing.

Three gaps matter more than the label. Coverage: many products publish clients and keep the server side closed, so "open source" can describe a quarter of the system you depend on. Reproducibility: unless builds are reproducible, the binary in the app store cannot be checked against the published source, and most are not. Review: published code that nobody audits is a possibility of scrutiny, not scrutiny. Closed source is not automatically worse — an audited closed-source product with a public report can be better evidenced than an unaudited open one. Treat this as one input to trust, read alongside audit history and who owns the company.

Read the specifics, because "supports passkeys" covers several different levels: creating passkeys in the manager, syncing them across your devices, using them in a mobile browser (which needs OS-level autofill integration and lags desktop), and exporting them out again. Cross-manager portability arrived late and is still uneven, so passkeys can be the hardest part of your vault to move. The reason to want a manager holding them rather than Apple, Google or Microsoft is the same reason you left browser password storage: your passkeys work across platforms and are not tied to one ecosystem's account. Also treat the provider's coverage claim as separate from the sites' — many services still only accept a passkey as a second factor, not instead of a password.

The value is specifically in eliminating password reuse — a large share of successful account takeovers rely on the same credentials working across multiple sites. It doesn't protect against every attack (phishing, malware on the device itself, a compromised master password), which is why two-factor authentication on the vault remains a separate, additional layer.

It breaks IP-based correlation only. A site that has already set a cookie, has you logged in, or fingerprints your browser follows you straight through an address change, so rotation does nothing about the tracking most people actually encounter. It can also work against you: rotating mid-session logs you out of services that bind a session to an address, and triggers CAPTCHAs and fraud checks. Worth distinguishing from a dedicated IP, which is the opposite trade — a stable address that is easier to allowlist and easier to associate with you.

The point is what it does to a stolen vault file: without the key, a copied vault cannot be attacked by guessing the master password, because guessing is no longer enough. That closes the exact gap a provider-side breach opens, where the master password is the only thing standing between an attacker and the data. The cost is recovery — lose both the key and your access, and there is no reset, only whatever emergency kit or family/team recovery the provider designed. Managers without this model compensate with a strong key-derivation function and high iteration counts, which raises the cost of guessing rather than removing the option; treat the derivation settings as the thing to compare when a secret key is absent.

Anything you exclude from the tunnel is exposed exactly as it would be with the VPN off — your real IP, your ISP's DNS resolver, your ISP's view of the traffic. Availability is uneven: on most providers it exists on Android and Windows, is rarer on macOS, and is usually absent on iOS because of platform restrictions. Check whether the implementation splits by app or only by IP range or domain, since app-based splitting says nothing about where an app's background requests go.

Zero-knowledge (client-side) encryption means the vault that leaves your device is already encrypted — a breach of the provider's servers exposes encrypted blobs, not plaintext passwords, provided your master password itself is strong and not reused elsewhere. A weak or reused master password remains the single biggest way this model fails.

Every claim here is perishable, which is why it belongs with a test date and a named service rather than a checkmark. Platforms detect VPNs by identifying IP ranges registered to hosting providers, so the contest is the provider rotating addresses against the platform blocking them, and results differ per service and per server location on the same subscription. Two things matter more than a long logo list: whether the provider publishes or supports specific streaming servers, and whether support will tell you which ones currently work. This is a convenience feature, not a privacy one — unblocking capability says nothing about logging, jurisdiction or leak protection, and a provider marketing mainly on streaming access is worth checking harder on all three.

Two distinctions carry most of the weight. First, diagnostic telemetry (a crash stack, a startup time) is not the same as behavioural telemetry (which sites you visited, which items you opened), and products often describe both with one word. Second, off by default is a much stronger position than a settings toggle, because a toggle only helps the minority who find it. Watch for the connections that are not labelled telemetry and stay on anyway: update checks, safe-browsing lookups, extension-store pings and bundled search suggestions all reach a server on their own schedule. Privacy-focused forks such as LibreWolf and Mullvad Browser exist largely to strip these out of upstream builds.

The label covers policy, not capability. Two things matter beyond the yes: whether port forwarding is available (without it, you connect to peers but accept no incoming connections, which limits you to leeching on many trackers), and what happens to a copyright complaint the provider receives — a provider with nothing to hand over answers differently from one keeping connection records. Providers that allow P2P on a handful of locations concentrate that traffic, which is a common cause of the slow speeds people blame on the protocol.

The methods are not equivalent. A hardware security key (FIDO2/WebAuthn) resists phishing because the key checks which site is asking; a TOTP app does not, and SMS is weakest of all because a number can be moved to someone else's SIM. Note what 2FA does and does not cover on a zero-knowledge password manager: it guards server-side login, so it stops someone reaching your vault through the provider, but an encrypted vault file copied from the provider's storage is decrypted with your master password, and 2FA has no part in that. Which is why master-password strength and the key-derivation settings behind it still matter after 2FA is on. Storing your 2FA codes in the same manager as your passwords also collapses two factors into one — keep the manager's own second factor elsewhere.

WireGuard's smaller codebase and modern fixed cryptography (ChaCha20-Poly1305, Curve25519) generally give it faster speeds and better battery life than OpenVPN, but OpenVPN's longer track record and configurability still matter in some network environments. IKEv2 has been dropped by several major providers in recent years.

Each hides different things from different watchers: a plain proxy hides your IP from the destination site but not from your ISP; a VPN hides your traffic from your ISP but the VPN provider itself can see it; Tor removes any single party's ability to see both who you are and what you're visiting, at a real cost to speed.

Encryption alone doesn't make one VPN as trustworthy as another — the provider running the server can see your traffic the way your ISP normally would, which is why jurisdiction, no-logs audits, and the provider's own track record matter as much as the protocol used.

Incoming mail from outside arrives unencrypted and is encrypted on receipt, so the provider handles the plaintext for a moment before storing it — the guarantee covers storage, not the whole path. Message metadata is the bigger gap: addresses, timestamps and, on most providers, subject lines stay unencrypted because the server needs them to sort, search and deliver. That means a court order can still yield who corresponded with whom and when, which is often the information being sought. Also check what a password reset does: if the provider can restore your mailbox after you forget your password, some key escrow exists; if it cannot, a lost password means lost mail, and that trade-off is the honest sign the model is real.