End-to-End Encryption

When you send a message via WhatsApp or Signal, it doesn’t travel directly to your friend. It runs through computers belonging to the company that operates the service. With end-to-end encryption, the message is already turned into unreadable gibberish on your phone. Only the recipient’s phone can turn it back into readable text. So the company only sees the gibberish on its computers, not the text. The two “ends” in the name are the two devices; everything in between remains blind.

Why the provider isn’t supposed to be able to read it

Many services only encrypt messages on their way to the server. There, they are decrypted, stored, and re-encrypted for forwarding. This protects against outsiders on the WiFi network, but not against the provider itself. Anyone with access to these servers can read everything in plain text. This is exactly the gap that end-to-end encryption closes.

This is more than a technical nuance. A server full of readable chats is an attractive target for attackers. If it’s hacked, millions of conversations are exposed. With genuine end-to-end encryption, attackers only find useless data junk there. Authorities also cannot demand content from the provider that the provider itself doesn’t know.

That’s why the method is politically controversial. Law enforcement argues that it makes criminals invisible. Experts counter that a backdoor built in for the police ultimately stands open to everyone. A vulnerability doesn’t distinguish who uses it. This debate keeps resurfacing in the EU, for instance under the buzzword chat control.

Two keys per person

A padlock is a good image for this. Every user has an open lock that they distribute freely, and the only matching key, which they never hand out. The lock is called the public key, the key the private key. Anyone who wants to write to you locks the message with your lock. Afterward, only your private key on your device can open it.

The private key never leaves the phone. It is generated directly on the device when the app is set up. The public keys, on the other hand, are exchanged between devices via the server. So the server distributes locks, but it doesn’t possess a single key. Good messengers also change the keys continuously, often for every single message.

A typical misconception: only the content is encrypted. The provider can still see who wrote to whom and when. This accompanying data is called metadata, and it often reveals surprisingly much. Encryption is likewise irrelevant if your device itself is compromised. Malware on the phone simply reads the text straight off the screen.

Messengers, cloud, and the dispute over AI features

The method is best known from messengers. Signal and WhatsApp use it by default, Telegram only in special secret chats. Video calling, password managers, and some cloud storage also work this way. Apple's iMessage and FaceTime are likewise end-to-end encrypted. In apps, this is usually indicated by a notice or a padlock icon in the chat window.

Many apps offer a security number or a QR code for comparison. This lets you check whether it’s really your counterpart’s device on the other end. Without this check, someone could theoretically position themselves in between and distribute their own locks. In practice hardly anyone does this, but it’s still worthwhile.

In the news, the term currently comes up mainly in connection with AI. A chat assistant that summarizes messages needs to be able to read them. If it runs on a server belonging to the provider, encryption is broken at that point. Companies are trying to solve this by having the AI compute directly on the device or in especially shielded data centers. How well such promises hold up is one of the industry’s open points of contention.