Browsing the internet safely
Browsing safely comes down to a few small habits whose effects are wildly unequal: keeping the browser updated is worth more than everything else put together, and after that, cutting down your extensions and being careful with what you download. The padlock beside the address bar has no important place on that list, because it only says the connection is encrypted and says nothing at all about whether the site is honest.
- Lesson 10 of 12
- Beginner
- Free, no signup
What actually works and what only feels like security
Has a real effect
- automatic browser and OS updates, switched on
- deleting extensions unopened for a year
- file extensions made visible in Windows
- taking software from the maker own site
- taking certificate and download warnings seriously
Not security on its own
- seeing the green padlock and relaxing
- clearing history regularly
- stacking several security extensions
- living in a private window
- trusting the top rank of a search result
The second column is not a list of wrong moves; they are moves that do not make you safe on their own and often take the place of the first column.
Last checked: Facts and tool names in this lesson are re-checked against their sources on this date.
Which habits actually work, and in what order?
First of all: keep the browser updated. That sentence has been repeated so often that it is no longer heard, but repetition does not change the order of importance. Almost every hole that reaches your device merely by opening a page is a hole in the browser itself, and its patch shipped weeks before you heard about it. A browser three versions behind is more exposed than any security setting you can put on top of it.
The real order, by what each habit actually stops, is this: update the browser and the operating system; cut down the extensions; be careful with downloaded files; use a different password for accounts that matter; and last of all, the thing most talked about, looking at the padlock and at the address.
Two things that are not on the list and feel like they should be. First, clearing history regularly: it is not a bad habit, but it is not security, it is privacy, and it has its place in the privacy lesson. Second, installing an extra security tool into the browser: most of them are themselves an extension, and the third section of this lesson is about extensions.
Our position: if you are going to do exactly one thing, turn automatic updates on and leave them alone. If two, delete the extensions you have not opened in a year. The rest of the list is smaller than those two.

What does the padlock beside the address bar prove?
One thing, and only that one thing: your connection to that domain is encrypted and nobody in the middle can read it or change it. About who sits behind that domain and whether they are telling the truth it says nothing, and it was never meant to. The technical detail is in the lesson on HTTP and HTTPS.
The issuers say the same thing themselves. Let's Encrypt, whose free certificates sit on millions of sites, wrote in its own explanation that a domain validation certificate asserts only that a public key belongs to a domain and "says nothing else about a site's content or who runs it", and adds plainly that such a certificate carries no information about a website's reputation, real-world identity or safety.
The certificate of the very site you are reading is a good specimen. Open it, and the subject field holds exactly one thing: the domain name. No company name, no registration number, no country. Precisely what a fraudulent site can also obtain in a few minutes at no cost at all.
And a real specimen from this site own contact form, on the day this lesson was being written: a message arrived pushing the reader towards a page about a Bitcoin transfer. That page was hosted on a completely legitimate and well known text publishing service, with sound HTTPS, a green padlock, and a domain name nobody would suspect. None of the old tells applied. What gave the message away was something else entirely: a cryptocurrency transfer nobody had asked for, arriving through a form meant for web design enquiries.
So what is the right tell? Not the padlock, but coherence. Is this message the answer to something you started? Is the domain in the address bar the place you think you are, letter by letter? Does what is being asked of you match what that site actually does? If all three are yes, the padlock is a bonus. If not, the padlock rescues nothing.
What the padlock says and what it does not
It does not prove
- which company or person sits behind the domain
- that the site does what it promises
- that the file it hands you is clean
- that the domain is not an imitation of another brand
It proves
- your connection to that domain is encrypted
- nobody in the middle can alter the content
- the public key belongs to that same domain name
The absence of a padlock now carries another meaning too: a site still without HTTPS is usually neglected, and neglect is a risk of its own.
Why is a browser extension the most overlooked risk?
Our position, said plainly: a browser extension has the widest gap between how risky it is and how much attention it gets. Most people think for an hour before installing an app and install an extension in two seconds, while the extension sits in the one place their entire online life passes through.
The reason is written in the permissions line and almost nobody reads it. A sentence like "read and change all your data on all websites" means exactly what it says: that extension can see the contents of your bank page, can read what you type, and can change the page so that you do not notice. Google own documentation for extension authors classifies permissions with exactly this bluntness, and Mozilla has a page saying what each permission request message actually means.
The second problem, and it is the worse one: extensions change hands. A simple, harmless extension that has grown popular over a few years can be sold, and the new owner can add whatever they like in the next version. You trusted something that is itself able to change, and automatic updates deliver that change to your device without a sound.
We live the same thing at a different scale on our own server. This site runs twenty four active WordPress plugins, and each of them, exactly like a browser extension, executes with the same authority WordPress itself has. That is why we keep the count down, why we wrote most of them ourselves, and why a plugin whose job is finished stays deactivated rather than left on just in case. The rule is the same in both places: less code, less attack surface.
The practical move: open your browser extensions page, delete anything you have not opened in a year, and read the permissions column for the rest. A unit converter does not need to read and change data on all websites.
How do you tell what a file you downloaded really is?
Not from its extension. We say that not from an article but from our own code. Our upload layer looks at the file extension first, but does not stop there: it then reads the real content of the file and rejects it if the content does not match the extension, and the message the user sees says exactly that. For images it goes one step further and tries to actually open the file as an image; a file with a picture extension that is not a picture is caught precisely here.
The reason is simple: a file extension is a claim by whoever named the file, not a fact about its contents. Windows also hides known extensions by default, and that one setting is what displays a file called "invoice.pdf.exe" on your screen as "invoice.pdf".
Four habits with the most effect. First, turn on the display of file extensions in Windows settings; once, forever. Second, take software from the maker own site rather than the first search result, because the adverts above the results are exactly where poisoned builds sit. Third, be suspicious of a file you were not looking for that arrived on its own; email attachments and files sent in messengers are the main route malware takes to an ordinary user. Fourth, when the operating system warns you as you open a file, take that warning seriously; getting past it needs one more click, and that click was put there deliberately.
Installing safely and checking that a file is what it claims has a lesson of its own: installing software safely.
Is public wifi still dangerous?
Far less than it was ten years ago, and less than what is still written about it. That old picture, somebody sitting in a cafe plucking your banking password out of the air, was built on a world where most sites had no encryption. Today nearly everything you open is encrypted and that attack no longer works easily.
But "less" is not "none", and what remains usually goes unsaid. The figure below shows the four real layers. The owner of the network still sees which domains you connect to, even when they cannot see the contents; the domain name is still visible in the early stages of the connection and in the name lookup. The page that opens when you join and asks for a password or a phone number is not the internet, it belongs to whoever runs that network. If your browser is out of date, the network you are sitting on is one of the best places to exploit exactly that hole. And the last one has nothing to do with networks: the person standing behind you can see your screen.
So in practice? Four things. Turn off automatic joining of open networks on your phone, because your phone will otherwise attach itself to any network with a familiar name. Turn off file sharing on a public network; Windows asks you this as you connect and most people click past it. If the browser warns you about a site certificate, take it seriously on a public network in particular. And do the things that involve money or identity over your own phone connection rather than the hotel free wifi; not out of fear, but because it costs nothing.
And honestly: for most people the bigger risk is not airport wifi, it is the extension they installed three years ago and forgot.
Four things that still remain on a public network
-
Which domains you opened
The content is encrypted, but the domain name is still visible in the name lookup and at the start of the connection.
-
The join page is not the internet
The page that opens as you connect belongs to whoever runs that network. Handing it a number or a password is handing them to a stranger.
-
An old browser on a strange network
This is where an unpatched browser costs the most, because the network is in somebody else hands.
-
The person standing behind you
Nothing to do with the network, and it rarely appears on a list of public wifi risks, yet it is more common than the rest.
None of these four means "they get into your account". Their real effect is smaller and more specific than what is usually said.
The fast path, with AI
The ordinary way to review your extensions is to look at the list and, because they all seem harmless, delete none of them. The fast way is to pull out the text of their permissions rather than their names, and ask a model to compare each extension against the job it claims to do. What surfaces in a few minutes is the thing that has been overlooked for years: an extension whose permissions do not match its purpose.
- Open the browser extensions page. In Chrome the address is chrome://extensions and in Firefox about:addons; open the details section of each extension to see the full permission text.
- For each extension write three things on one line: the name, one sentence about what it does, and the permission text. Those three columns are the model input.
- Give the recipe below to a fast model together with that list. The job is comparison rather than judgment, so a cheap model is enough here and even a few dozen extensions finish quickly.
- For every extension the model flagged, open its store page yourself and look at three things: the date of the last update, the name of the developer, and whether ownership has changed. Then decide.
Copy-ready recipe
I am giving you the list of my browser extensions. Each row has three columns: name, the job it claims, and the permission text.
For each row build a table with these columns:
1. Extension name.
2. Are these permissions necessary for that job: necessary, excessive, or unclear.
3. If excessive, which permission is the surplus one.
Rules:
- Base the answer only on the permission text I gave you. Write nothing about the extension itself from your memory.
- Count "read and change all your data on all websites" as excessive for tools that work on one site only.
- If the description of the job is unclear to you, write unclear and say why in one sentence.
At the end, list the excessive extensions separately, with no recommendation to delete anything; the decision is mine.
The list:
{name, job, permissions}
Before you trust the output: The model does not know who owns an extension or what it will do in the next version; it only weighs the two texts you supplied against each other. So its output is a list of things worth looking at, not a verdict. The final decision comes from the store page: who the developer is, when the last update was, and whether ownership has changed. And paste the permission text as it stands; if you summarise it yourself, the very sentence that revealed the problem disappears.
AI in this kind of work
For this topic a language model does one specific job well: reading the tedious texts nobody reads, meaning extension permission lines and terms of use, and showing where they fail to line up. The job it must not be given is this: deciding that a site or a file is dangerous. The model neither opens the site nor scans the file, and its answer comes only from the surface shape of a name and a text.
Tools that actually help
- Gemini Enough for comparing a few dozen permission lines against a description of the job, and a fast model beats an expensive one here because the work is repetitive. Google own page says the Gemini app works in more than 230 countries and territories, and Iran is not on that list.
- Claude For when the terms or the policy of an extension run to several pages and you want to know where permission to sell data was taken. Iran is not on Anthropic list of supported countries, so there is no official route to sign up or pay.
- ChatGPT It serves the same comparison job and is not meaningfully different from the others. On reaching it from Iran we write nothing: the OpenAI supported-countries page returns 403 to this server, so we have not checked it and we do not make a claim with nothing behind it.
Where it backfires
The specific risk of this lesson is this: the AI assistants that install into the browser are themselves extensions, and in order to read and summarise the page you are looking at they ask for precisely the permission this lesson warns about, that is, reading and changing data on all websites. Google own documentation for extension authors puts that permission at the highest warning level, and Mozilla has a page telling users what each permission request message means. This does not mean those tools are bad; it means the standard for choosing one should be stricter than for an ordinary extension, not looser. There is a further layer that is a genuinely new risk and has its own place: an assistant that reads pages on your behalf also reads any text on that page, including text somebody put there deliberately for that assistant. That subject is opened by this same security track, in its lesson on the security risks of AI.
Sources: Chrome for Developers: declare extension permissions Mozilla: permission request messages for Firefox extensions Google: where Gemini Apps are available Anthropic: supported countries
Where this advice stops
Everything in this lesson stands on one assumption: your device is clean. If something is installed on the computer or the phone that should not be, none of these habits work, because that thing sits in front of the browser and also composes what you see. The second boundary is that this lesson is about the user and not about a website: if you want to secure your own site, that is an entirely different matter and website security is where it belongs. And the third boundary, where no technical habit helps: if somebody talks you into handing over a password or a code by phone or message, no update and no extension stops it. That subject sits elsewhere, and its remedy is not technical either.
From our own work
Three things from this site. First, the TLS certificate of this very domain: its issuer is Google Trust Services, it was issued for three months and renews automatically, and its subject field holds exactly one line, the domain name. No company name, no country, no registration number. That is, the same certificate a fake site can also obtain, and precisely for that reason the padlock beside our own address proves nothing about us. Second, the upload layer of our own freelance platform: in that code, after the extension check, the real content of the file is read and the file is rejected if the content does not match the extension; the Persian message the user sees says exactly that the file content does not match its extension and it was not accepted. For image formats there is a further check that tries to actually open the file, and that is what catches a non-image with an image extension. Third, on the same day this lesson was written, a spam message arrived through this site contact form pushing the reader to a page about a Bitcoin transfer, and that page was hosted on a completely reputable text publishing service with sound HTTPS. We do not print its address here, but that one specimen is enough to show that a padlock and a reputable domain say nothing about the intent of the page.
Real follow-up questions
Is the first Google result safer?
The rank of a result says nothing about whether it is safe, and for the advert above the results not even that much holds, because that place is bought rather than ranked. The real danger sits right there: poisoned builds of a popular program usually sit in the advert above the results, under a name one letter away from the real one. To download software, type the maker site name straight into the address bar and do not go through a search result.
Does a site without HTTPS mean it is dangerous?
It means the connection is unencrypted, which is a different problem: anybody on the path can read and alter it. It does not say whether the site is dangerous. But today, when a certificate is free and issued in minutes, its absence carries a second signal: nobody has touched that site in a long time, and neglect is a risk in itself.
When the browser warns and offers a continue button, can I go on?
That button is deliberately hidden behind one extra click, which is the browser asking you to pause. If the warning is about a certificate and you are on a public network, do not go on; leave what you were doing there. If it is on your own network for a site you know and whose certificate you know has expired, the risk is lower, but never go past it on a page that asks for a password or a payment.