Encrypted Messaging Software: What "Encrypted" Actually Means and What It Doesn't

You're shopping for encrypted messaging software. Every app on the market says it's encrypted. Most of them are telling you something true, but almost none of them are telling you the whole truth. "Encrypted" is a spectrum, not a checkbox. The difference between the top and bottom of that spectrum determines whether your messages are actually private or just feel that way. This piece is a technical walkthrough of what encryption guarantees, what it leaves exposed, and how to evaluate the claims you'll encounter.

Key Takeaways

• End-to-end encryption (E2EE) means even the platform can't read your messages. But many apps that say "encrypted" only encrypt data in transit or at rest on their servers, which means the company can still access it.
• Encryption protects content. It usually does not protect metadata: who you talk to, when, how often, and from where. Metadata alone can reconstruct most of what a surveillance operation needs.
• The legal environment matters as much as the protocol. Governments in the UK and EU are actively pressuring platforms to weaken or bypass encryption. Where a service is domiciled determines what legal demands it faces.
• Disclosure records (what a platform actually produces under subpoena) are more honest than marketing copy. Check those instead of landing pages.
• Default-on encryption is the only kind that matters at scale. Opt-in encryption protects almost nobody, because almost nobody opts in.

What Does "Encrypted" Actually Mean?

Encryption is a mathematical process that turns readable data into unreadable ciphertext. Only someone with the correct key can reverse it. That part is simple. What varies, wildly, is where the encryption happens, who holds the keys, and what data never gets encrypted at all.

Three layers show up in messaging products, and they are not equivalent:

• Encryption in transit. Your message is encrypted between your device and the server. The server decrypts it, processes it, and re-encrypts it before sending it to the recipient. The platform can read the plaintext at rest on its infrastructure. This is the baseline for any HTTPS connection. It stops a coffee-shop attacker. It does not stop the platform itself.
• Encryption at rest. The server stores your data in encrypted form. This protects against certain breach scenarios (an attacker who steals a disk image, for example). But if the platform holds the decryption keys, it can still read the data whenever it wants. "Encrypted at rest" is a storage property, not a privacy guarantee.
• End-to-end encryption (E2EE). Your device encrypts the message with a key that only the recipient's device can decrypt. The platform never sees plaintext. It can store ciphertext, route it, and deliver it, but it cannot read it. This is the only architecture where "we can't read your messages" is a structural claim rather than a policy promise.

When a product says "encrypted," you need to determine which of these three things it means. Most don't volunteer the distinction.

Why Does Default-On Encryption Matter More Than Opt-In?

Because opt-in encryption protects a rounding error of users. The pattern repeats across platforms: a feature exists, almost nobody activates it, and the platform points to its existence as proof of privacy commitment while the vast majority of conversations remain readable.

Telegram is the canonical example. Its "Secret Chats" feature uses end-to-end encryption. Regular chats, group chats, and channels do not. They sit on Telegram's servers in a form the company can access. Since Secret Chats must be manually activated per conversation and don't work for groups, the overwhelming majority of Telegram traffic is not end-to-end encrypted. The marketing says "encrypted." The architecture says "mostly not."

Meta's Instagram followed a similar arc. It quietly introduced end-to-end encrypted DMs in December 2023, made them opt-in, observed low adoption, and is now discontinuing the feature entirely as of May 8, 2026. The lesson is clean: opt-in encryption that nobody uses gets removed.

Compare this to Signal, which encrypts every conversation, call, and group chat by default. There is no unencrypted mode. The Signal Protocol is open source, and other apps (including WhatsApp) have licensed it. But "uses the Signal Protocol" and "uses the Signal Protocol by default, with no way to turn it off" are different claims with different privacy outcomes.

What Is the Signal Protocol and Who Uses It?

The Signal Protocol is a cryptographic specification for end-to-end encrypted communication. It handles key exchange (the Double Ratchet algorithm, built on X3DH), forward secrecy (compromising one key doesn't decrypt past messages), and post-compromise security (a future key rotation heals the session even if a key was stolen). It is open source and has been independently audited.

Signal (the app) uses it. WhatsApp uses it. Google Messages uses it for RCS conversations between two Google Messages users. The protocol is sound. But the protocol is only one component. What the platform does with everything around the encrypted payload determines your actual privacy posture.

What Does Encryption Not Protect?

Metadata. This is the gap that most comparisons skip, and it is large.

Even on a platform with perfect end-to-end encryption of message content, the provider can typically still observe: your account identifier (usually a phone number), who you communicate with (contact graph), when you communicate (timestamps), how often, message sizes, your IP address, and your device information. Group memberships are often visible. So are read receipts and typing indicators at the network level.

This is not a marginal concern. Metadata analysis can reconstruct social networks, identify sources for journalists, map organizational hierarchies, and locate individuals. A former NSA general counsel once said, "We kill people based on metadata." The content of your messages could be perfectly opaque and your communication patterns could still be fully transparent.

Different platforms handle metadata differently:

• Signal is architecturally minimalist. It uses sealed sender to hide the sender's identity from Signal's own servers. It collects almost no metadata and has demonstrated this under subpoena: when compelled to produce records, Signal has produced essentially nothing, because it has essentially nothing to produce.
• WhatsApp uses the same encryption protocol for content, but Meta can still see relationship metadata: who you talk to and when. It shares this metadata within Meta's product ecosystem.
• iMessage encrypts content end-to-end between Apple devices. But green-bubble SMS conversations with non-Apple phones are unencrypted, and Apple retains some metadata including contact lists and timestamps.

The label "end-to-end encrypted" does not distinguish a platform that engineers hard against its own visibility from one that encrypts content while retaining a rich metadata profile of every user.

How Do You Know What a Platform Actually Collects?

Disclosure records beat marketing copy. What a platform proves able to produce under legal compulsion is the most candid documentation of its architecture.

When law enforcement serves a subpoena or court order, a platform must produce whatever responsive data it has. If it has nothing, it produces nothing. Signal's transparency reports consistently show near-empty responses. WhatsApp's responses include subscriber information, metadata, and in some cases address-book contacts. Telegram's responses have historically included IP addresses and phone numbers.

This is a more reliable signal than any privacy policy. A privacy policy describes intent. A subpoena response describes architecture.

What Happens When Governments Try to Break Encryption?

They're trying right now, in multiple jurisdictions. The technical fight over encryption is largely settled (E2EE works). The legal fight over whether you're allowed to keep it turned on is active and accelerating.

The UK: Online Safety Act and Apple

The UK's Online Safety Act, which received royal assent in October 2023, grants Ofcom the authority to impose measures against illegal content on messaging platforms. In practice, this creates a framework where regulators can demand platforms scan encrypted messages for prohibited material.

Apple has already been forced to weaken UK security once. Rather than build a backdoor into iCloud encryption, Apple withdrew its Advanced Data Protection feature from UK users entirely, leaving UK residents with less encryption than users elsewhere. A second legal fight is now active: Apple is challenging a newer "technical capability notice" from the UK government demanding access to users' encrypted data.

Signal's president Meredith Whittaker has stated flatly that Signal would "absolutely, 100% walk" from the UK rather than comply with demands to weaken encryption. That statement was made in June 2026.

The EU: Chat Control

The EU's "Chat Control" proposal has gone through multiple iterations. The most aggressive version would have required platforms to scan encrypted messages for child sexual abuse material (CSAM). As of April 2026, EU lawmakers blocked the mass-scanning provision, though they permitted providers to resume voluntary scanning activities until April 2028 subject to safeguards. The forced scanning of encrypted messages was given up by EU member states, but the broader regulatory framework continues to evolve, with the real fight expected to continue into late 2026.

The technical objections are not political. Over 500 cryptography and security researchers have signed an open letter warning that these scanning measures are technically unfeasible and would undermine the security of all European citizens. You cannot build a backdoor that only the good guys can use. A vulnerability introduced for law enforcement is a vulnerability available to anyone who discovers it.

Why This Matters for Your Choice of Messaging App

Where a service is domiciled determines what legal regime it faces. A company headquartered in a jurisdiction with active backdoor mandates may be compelled to weaken encryption regardless of its current marketing claims. The question is not just "is it encrypted today?" but "will it still be encrypted next year, and does the company have a public commitment about what it will do if ordered to compromise?"

Signal has made its position explicit: exit rather than comply. Apple has shown a mixed record: withdrawal of features in one case, legal challenge in another. Meta has shown willingness to remove encryption features when adoption is low. These are different risk profiles.

How Should You Evaluate Encrypted Messaging Apps?

Here is a framework. No scoring system, no "best of" ranking. Just the questions that actually determine your privacy outcome.

1. Is encryption on by default? If you have to opt in, assume it's off. For you and for everyone you talk to.
2. What metadata does the platform retain? Check disclosure records, not marketing pages. If the platform has never received a subpoena or has never published a transparency report, that is itself a data point.
3. Is the protocol open source and audited? Closed-source encryption requires you to trust the vendor's implementation without verification. Open-source protocols (like the Signal Protocol) can be independently reviewed.
4. Where is the company domiciled? This determines what legal demands it faces. A company in the UK faces the Online Safety Act. A company in the EU faces evolving Chat Control regulations. A company in the US faces National Security Letters and FISA orders.
5. What has the company said publicly about government demands? Look for explicit commitments, not vague privacy language. "We will exit a market rather than weaken encryption" is a different statement than "we take your privacy seriously."
6. Does the app protect content only, or content and metadata? This is the dividing line between good and exceptional privacy engineering.

What About Encryption in AI Products?

The same questions apply, and almost nobody is asking them yet.

AI assistants process your prompts, your documents, your personal context. The privacy surface is arguably larger than messaging, because you're sharing not just conversations but work product, personal history, and behavioral patterns. Yet the scrutiny applied to messaging encryption is largely absent from AI product evaluation.

The questions map directly. Is the data encrypted at rest? Is it encrypted end-to-end, or does the provider see plaintext at inference? What metadata is retained (timestamps, usage patterns, prompt topics)? Where is the company domiciled? What happens to your data if the company receives a legal demand?

For most AI chat products, the answers are unflattering. Prompts are typically sent in plaintext to inference servers. Conversation history is stored on provider infrastructure. Metadata is retained indefinitely. The encryption posture is comparable to Telegram's default mode: encrypted in transit, readable by the platform.

We built Selina with this problem in mind. Memory is encrypted at rest, and files and transfers through SelinaSEND are end-to-end encrypted. But memory is not end-to-end encrypted, because a slice of each request reaches a frontier provider at inference. We state that because it's true, and because you should hold AI products to the same standard you'd hold a messaging app.

What Is Changing at the Protocol Level?

One development worth tracking: RFC 9849, published in March 2026, specifies TLS Encrypted Client Hello (ECH). This allows a client to encrypt sensitive portions of the TLS handshake, including the server name indication (SNI) field. Previously, even on an encrypted connection, a network observer could see which specific service you were connecting to. ECH closes that gap at the network layer.

This is infrastructure-level progress. It doesn't change application-layer encryption (your messages are still encrypted, or not, by the app itself). But it shrinks the metadata surface at the network layer, making it harder for ISPs and network-level surveillance to determine which services you're using.

How Do You Compare the Major Apps Side by Side?

Flat summary of the current state, based on architecture rather than marketing:

• Signal. E2EE on by default for all messages, calls, and groups. Open-source protocol. Minimal metadata collection, demonstrated under subpoena. Nonprofit foundation. Has publicly committed to exiting jurisdictions rather than weakening encryption.
• WhatsApp. E2EE on by default using the Signal Protocol. But operated by Meta, which retains relationship metadata and shares it across its product ecosystem. Content is protected; your social graph is not.
• Telegram. E2EE only in opt-in Secret Chats. Regular chats are cloud-encrypted with keys Telegram holds. Groups and channels are not E2EE. The gap between marketing and architecture is wide.
• iMessage. E2EE between Apple devices. Falls back to unencrypted SMS for non-Apple recipients. Advanced Data Protection for iCloud backups has been withdrawn for UK users under government pressure.
• Meta Messenger. Rolled out default E2EE in December 2023. Still active. But Instagram DMs are losing their encryption feature in May 2026.

What Should You Actually Do?

If message content privacy is your primary concern and you want the strongest architecture available today, Signal is the clear choice. This is not a controversial position among security researchers. It is approximately the consensus position.

If you need to communicate with people who won't install Signal (which is most people), WhatsApp provides strong content encryption with weaker metadata protection. That's a real tradeoff, not a failure.

If you use Telegram, know that your non-Secret-Chat messages are not end-to-end encrypted. If you use iMessage, know that your green-bubble conversations are not encrypted at all. If you used encrypted Instagram DMs, download your data before May 2026.

And if you're evaluating AI products (or any product that processes personal data), apply the same framework. Ask what is encrypted, where the keys are, what metadata is retained, and what the company will do when a government asks for a backdoor. The answers matter more than the label.

If you want an AI assistant that takes this seriously, start a free 7-day trial of Selina, no card required.

Michael C.

Founder & Principal Engineer, Selina Labs

Michael builds Selina, a privacy-first AI that remembers you across conversations. He ships security-sensitive AI in production — real attacks, real fixes, measured in minutes and dollars — and writes about privacy, security, and LLMs from that seat. Top Rated Plus and expert-verified on Upwork.

• www.upwork.com
• x.com
• medium.com
• selinalabs.io