How can you choose a web browser without built-in AI features?

What “without AI” can mean

The observed Hacker News question explicitly rejects a browser that merely offers a button to turn AI off. That is a valid preference, but it needs a testable definition. “AI” is used for very different things, from a cloud chatbot that reads a tab to local translation, image descriptions, ranking, autocomplete, spellchecking, fraud detection, and accessibility features. Mozilla itself distinguishes newer generative features from conventional machine-learning functions such as classification, ranking, and personalization. Mozilla support documentation

Use the following vocabulary when comparing browsers.

This table prevents a common mistake. A browser can have no visible chatbot while still including on-device generative features. Conversely, a browser can contain an AI menu but never send page data to an AI service unless the user enables it. The privacy, cost, and philosophical implications are different, so write down which of them you are rejecting.

What the available evidence says about each option

The current evidence supports these categories as of 2026-09-04. They are not a promise about a later release, a distribution package, a mobile version, or a particular region.

For the strict absence requirement, the opt-out options in this table do not qualify. A smaller browser may be worth investigating, but its feature page alone cannot establish what every dependency contains. Auditing a custom build also means keeping that build patched.

Firefox is a strong opt-out, not an absence claim

Firefox is useful to people who do not want generative browser features but still want a well-supported daily browser. Its desktop AI Controls panel, available from Firefox 148, covers translations, PDF image alt text, AI-enhanced tab groups, link-preview key points, and the sidebar chatbot. Turning on “Block AI enhancements” hides current and future features and their promotional prompts. When a feature is blocked, Mozilla says that downloaded on-device models for it are removed. Mozilla support documentation

That setting does not make the program an AI-free binary. Mozilla explicitly excludes traditional ML from this panel, and it says independently installed extensions can still use third-party AI services. It also lists search engines, website features, address-bar behaviour, and New Tab personalization as separate settings areas. Mozilla support documentation Choose Firefox with the block enabled when your practical rule is “do not expose me to Firefox generative features,” not when your rule is “no AI-related code may ship.”

Chrome is an opt-in example, not an AI-free answer

Google describes Gemini in Chrome as a gradually released feature that requires a supported region and device, the latest Chrome, and a signed-in profile. It says users opt in the first time they use it. If it is enabled, Gemini can use the current tab by default and, on desktop, can use up to ten open tabs when they are shared. Chrome provides settings for its visibility and permissions, including current-tab sharing and browsing permission. Google Chrome Help

Those controls can reduce use, but they do not satisfy an absence test. The same distinction applies to any browser whose assistant is unpinned, hidden, signed out, or disabled. A visible switch is valuable for consent, but it proves that the capability exists.

Treat smaller and privacy-focused browsers carefully

Falkon's official site currently presents a conventional feature set: an ad blocker, password manager, themes, extensions, Plasma integration, vertical tabs, and release notes focused on ordinary browser changes. It does not present a generative assistant in the material reviewed. Falkon project site That makes it a sensible candidate to test if you use Linux or KDE and want a browser whose public feature set is not organized around AI. It does not prove the stronger proposition that there is no machine-learning-related dependency or no future AI feature.

LibreWolf and Mullvad Browser illustrate why privacy and AI are separate axes. LibreWolf publishes configuration defaults that disable telemetry and several remote services, but its project also depends on a changing browser upstream and its official development location has moved. LibreWolf project notice Mullvad Browser is designed to reduce tracking and fingerprinting and is signed by Tor Browser Developers, but its stated purpose is privacy rather than a permanent no-AI pledge. Mullvad Browser repository Use these products if their privacy or anti-fingerprinting model matches your needs, then verify their current AI-related behaviour just as you would any other browser.

How to check a browser before switching

1. Write a one-sentence policy

Examples make the choice concrete.

• “I do not want a browser-supplied generative assistant, summary, translator, image generator, or tab organizer, but I accept ordinary spellcheck and local security heuristics.”
• “I do not want browser content sent to a cloud AI service, but local accessibility functions are acceptable.”
• “I will install only a build for which I can audit and exclude all model-related code and dependencies.”

If the first policy means those features must be absent, a block setting is insufficient. If it means they must not run, a documented block may be acceptable. The second policy needs a network and account review. The third may require building from audited source, checking dependencies, and maintaining security updates yourself.

2. Check the exact release, not the brand

Read the official release notes, feature documentation, privacy notice, and configuration documentation for the version and platform you will actually install. Mobile, desktop, stable, beta, enterprise, and distribution-packaged versions may differ. Look for terms such as AI, assistant, chatbot, Gemini, Copilot, model, summarize, translate, alt text, smart tab, link preview, semantic search, and on-device.

If a project claims no AI, ask what that means. Does it exclude only a cloud chatbot? Does it disable features by preference? Does it remove model downloads? Does it exclude all ML libraries? Is the claim about an official package, a fork, or a build option? If the answer is vague, classify it as unverified rather than absent.

3. Install in a fresh profile

Use the official download route and verify signatures or hashes where the project offers them. Do not start by importing a profile, signing into a browser account, restoring extensions, or accepting every first-run prompt. In the fresh profile, inspect settings, sidebars, context menus, PDF tools, tab groups, translation settings, New Tab content, address-bar suggestions, and permission prompts.

Then update once and repeat the check. A browser that meets your policy at first launch but changes after an update is not a durable solution. Keep automatic security updates unless you have a defined and prompt alternative patching process. Turning updates off to prevent new features can leave you exposed to known browser vulnerabilities.

4. Audit extensions and web apps

An extension can add an assistant to any browser, inspect page content, send selected text to a remote API, or modify search results. The Firefox documentation makes this explicit: blocking Firefox AI enhancements does not prevent extensions from using independent third-party AI services. Mozilla support documentation

Remove extensions you do not need. For the ones you keep, review the publisher, requested permissions, privacy policy, update history, and whether the extension calls an AI provider. Be equally cautious with sidebar web apps, pinned tabs, browser-account integrations, and “productivity” tools. An AI-free browser binary does not stop a web page from loading a chatbot.

5. Check search and new-tab integration separately

Search is a website or provider decision, not necessarily a browser decision. Google says AI Overviews are a core Search feature that cannot be globally turned off, though its Web filter shows text links without AI Overviews. Google Search Help Changing browsers alone will not remove an AI answer supplied by a search-results page.

Choose a default search provider whose current results and privacy terms fit your policy, and check the browser's address-bar shortcuts, visual search, query suggestions, sponsored suggestions, semantic-search settings, and New Tab page. Do not assume that a search provider is non-AI simply because you did not use a chatbot. Review it after major browser or search-service changes.

6. Use network checks as evidence, not as proof

For a sensitive use case, test the browser with a clean profile while you perform a small, repeatable set of actions: launch it, open a neutral local page, browse a few ordinary sites, open a PDF, create a tab group, use translation, and type in the address bar. A device firewall, DNS log, or approved network-monitoring tool can show unexpected connections to model or analytics services.

This check has limits. Encrypted traffic and shared content-delivery domains make endpoints hard to interpret, and an absence of observed connections does not prove a local model is absent. Do not upload private browsing content to a third-party “AI detector” or traffic-analysis service merely to perform the test. Record the browser version, operating system, enabled extensions, test steps, and observed destinations so you can compare after updates.

AI features, telemetry, and privacy are different questions

It is reasonable to care about all three, but do not use one as a proxy for another. A local feature can be AI without sending data off the device. A cloud feature can have clear opt-in and privacy controls but still be unacceptable to you. A browser can have no visible assistant yet collect telemetry or use a search provider whose website offers AI answers.

LibreWolf's published configuration, for example, explicitly disables its Firefox telemetry switches and certain recommendation and remote-settings features. LibreWolf settings source That can be useful privacy evidence, but it does not prove an all-AI-free state, and it says nothing about data a website receives when you visit it. Evaluate feature surface, network destinations, telemetry, and web-service choice as separate checklists.

Privacy-focused anti-fingerprinting browsers need an additional caution. Adding a distinctive configuration or many extensions can make you easier to recognize, which conflicts with their design goal. Tor Project guidance discourages extra Tor Browser add-ons because they can harm privacy and make a user easier to track. Tor Browser extension guidance That is a privacy rule, not an AI rule, but it shows why a browser's stated purpose matters.

Operating-system features can appear inside any browser

An AI-free browser does not control the operating system around it. On a compatible Apple device, Apple says Writing Tools can work nearly everywhere you type, including third-party apps. Apple support documentation That means a rewrite, summary, or composition command can appear while you are working in a browser even if the browser itself has no AI feature.

Review the operating system's AI, assistant, writing, accessibility, transcription, and text-selection settings separately. Check whether models have been downloaded and whether an external-provider extension has been enabled. Do the same on managed work devices, where an administrator can provide browser policies, extensions, search defaults, security software, or system assistants that differ from your own settings.

The same principle applies to passwords, spellcheck, translation, dictation, screen readers, and security software. Some are useful accessibility or safety tools; some use ML; some rely on cloud services. Do not disable a protective feature casually just because it is described with AI language. First determine whether it is generative, local, cloud-backed, optional, and necessary for your risk tolerance.

Example of choosing between absence and opt-out

Suppose you want no browser-supplied summaries, chat sidebars, or generated tab names. First decide whether you reject those features being included or only being enabled. This determines which options are worth testing.

If inclusion itself is unacceptable, do not start with an opt-out setup. Check a candidate release's feature documentation and build information, and leave its status unverified where the evidence is incomplete. Confirm that it works on your operating system and receives security updates.

If you accept included features provided they stay disabled, a documented block such as Firefox's may meet that looser requirement. Check its stated exclusions and review extensions and search separately. Describe the result as features blocked, so you do not mistake a configuration choice for proof about the browser's contents.

Tradeoffs worth making explicit

The security tradeoff is particularly important. “No AI” is not a substitute for a maintained browser, phishing protection, account security, backups, or healthy skepticism about web content. A browser you can patch promptly and configure consistently is usually safer than an abandoned fork that happens to have fewer visible features.

A short recurring audit

• After every major update, review release notes and search settings for new model, assistant, summary, translation, or recommendation features.
• Confirm your browser's AI or assistant controls remain blocked, if your browser has them.
• Review all installed extensions and remove anything you do not recognize or need.
• Check the default search engine, address-bar shortcuts, New Tab content, signed-in browser account, and sync state.
• Review operating-system writing, assistant, transcription, and accessibility features separately.
• If cloud AI is a hard boundary, repeat the clean-profile network test after feature updates and document what you checked.
• Keep security updates on, or move to another currently maintained browser rather than freezing an old version.