End-to-end encryption (E2EE)

basicwidespread

Encryption in which only the sender's and the recipient's devices can read the message, not even the intermediary service provider.

In plain terms

Imagine sending a letter in a locked box where only the recipient has the key. The postman delivers it, the post office stores it, but neither can look inside. End-to-end encryption does the same for your messages, calls, and files.

The 'endpoint' is your device and the recipient's, such as two phones. The message locks on your phone and unlocks only on the other. What passes through the provider's servers in between is an unintelligible string of code.

This is more than standard encryption, which only protects data in transit while the content remains readable on the provider's servers. Here, the provider itself cannot release or analyze your conversation because it lacks the key.

In the AI era, this raises a new question. If an AI assistant or agent reads and summarizes your messages, it must have access to the content, so it matters whether it works on your phone or on a remote server.

How it works

When installing the application, a key pair is generated on your device: a public key that anyone can obtain, and a private key that typically never leaves the device.
When sending a message, the two devices agree on a shared secret key using public keys in such a way that the intermediary server cannot calculate it. The recipient does not even need to be online for this.
Your device encrypts the message, and the server only forwards and stores the sealed packet until the recipient accesses it.
The recipient's device opens the package using its own private key and the shared key. Modern applications exchange keys per message, so a leaked key does not expose previous conversations.
Parties can verify they are truly communicating with each other, and that no one has inserted themselves in between, by comparing a security code or QR code.

An everyday example

When you send a photo to a family group on WhatsApp or Signal, only the devices of group members can display it. However, if you ask an AI assistant built into WhatsApp to summarize the conversation, the text must be visible to the AI, so it is worth checking where and under what protection the processing occurs.

Who is affected

everyoneThe privacy of your daily messages, photos, and calls depends on whether the application used encrypts them in this way.

entrepreneur, executiveThe protection of trade secrets and customer data, as well as the selection of corporate AI tools, depends on who has access to the content of communications.

patient, doctorHealth data is particularly sensitive; therefore, test results or complaints should only be sent via channels where even the intermediary cannot see the content. → Health

developerIf you build an AI function or agent on encrypted data, you must decide where the content becomes readable, and the user must be able to see this. → Code

A common misconception

MythIf an application uses end-to-end encryption, then no one can learn anything about my conversations.

In realityEncryption protects the content in transit and on the server, but not the endpoints. Anyone with access to your unlocked phone, unencrypted backups, or metadata (who, when, how often you talk) can still learn a lot.

Limits and risks

Protection ends at the endpoints: it does not defend against spyware, a stolen phone, or a recipient taking a screenshot.
Conversations saved to the cloud are often not encrypted in the same way, so check the save settings separately.
If you give an AI assistant or agent access to your messages, the content may leave the encrypted channel, especially if processing occurs on a remote server.
If you lose your key or your recovery code, the provider cannot return your data.

Similar, but not the same

encryption in transit (TLS, HTTPS) – It only protects the path between your device and the server; the content remains readable on the provider's server.
encryption of stored data – It protects data on the disk against theft, but the key is typically held by the provider, so they can access the content.
anonymity – End-to-end encryption hides what you say, but usually not who you are or who you are talking to.

Related concepts

Broader

encryption, data protection, cybersecurity

Narrower

Signal-protokoll, MLS-protokoll

On the edge

public-key encryption, forward secrecy, on-device AI, metaadat, two-factor authentication, confidential computing, client-side scanning

A brief history

The principle is based on the 1976 description of public-key cryptography (Diffie and Hellman). One of the first widely used implementations was Phil Zimmermann's PGP email encryption released in 1991. It became mass-market when WhatsApp made it default for all users in 2016, built on the Signal protocol.

Where things stand

As of late 2026, it is the default in most major messaging apps, such as WhatsApp, Signal, iMessage and Messenger. The debate revolves around two questions: whether authorities should be able to access encrypted content (for example, through mandatory message screening), and how an AI assistant or agent can work on it while the promise of encryption is preserved.

Status: 2026 November

In the news

Appears in 1 Agent Age stories; the curve shows weekly occurrence in the news magazine.

WhatsApp to screen suspected scam messages with on-device AIAug 31, 2026 · Money

Also known as

End-to-end encryption · end-to-end encryption · end-to-end encryption · end-to-end encryption · end-to-end encryption · E2EE · E2E

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.