The problem: “harvest now, decrypt later”
Most messengers agree on keys using elliptic-curve math (e.g. X25519). A classical computer cannot break it, but a quantum computer running Shor's algorithm can. Such a machine does not exist yet — and it doesn't have to: it is enough to record encrypted traffic now and wait. Conversations meant to stay private for years — personal, family, business — are already at risk today.
The solution: a hybrid of X25519 and ML-KEM-768
Szept combines two key-agreement mechanisms: the proven X25519 and the post-quantum ML-KEM-768 — a NIST standard (FIPS 203) designed to withstand quantum computers too. Both secrets go into the key derivation function (HKDF-SHA256) together, bound to everything the parties exchanged. To read a message you would have to break both — so the hybrid is never weaker than classical cryptography, even if a flaw were found in the new algorithm.
What exactly is protected
- Every new text conversation — including the first message to someone who is offline (their phone leaves a signed ML-KEM key on the server).
- All later messages in that conversation — every step of the Double Ratchet combines X25519 with a fresh ML-KEM-768 exchange, and each message key is deleted after use.
- Every voice and video call — gets its own one-time X25519 + ML-KEM-768 exchange and a separate key for each direction.
- Groups — a group message is encrypted separately to every member over the same sessions, and group-call keys are distributed through them.
Content is encrypted with AES-256-GCM on your phone. The Szept server only carries ciphertext — it does not have the keys and cannot have them.
Post-quantum re-keying on every exchange
Since Android 0.9.96, in Szept Web and on iPhone from version 0.2.1, Szept re-keys conversations post-quantum. On every change of direction in a conversation both sides perform an X25519 step and a fresh ML-KEM-768 exchange at once, and the new key is derived from both results.
What it gives you. If someone stole keys from a phone, the conversation “heals” — later messages become unreadable to them again. With ML-KEM in every step this also holds against an attacker with a quantum computer and recorded traffic, after a single exchange of messages.
How it differs from Signal. Since 2025 Signal has the post-quantum SPQR ratchet in a “sparse” form: the large ML-KEM key is split into pieces sent across many messages to save bandwidth. We send the full key and ciphertext in every message (about 2.3 KB) — unnoticeable on today's connections, and protection returns sooner. Signal, on the other hand, has a formally verified protocol and years of external audits; we publish the full protocol description and say so openly.
Compatibility. Older app versions keep talking to new ones without issues; the protection switches on by itself once both sides are up to date. Once on, it cannot be silently switched off — a message without the post-quantum part is rejected.
Szept compared with the most popular messengers
Of the six messengers in the table, only Szept re-keys post-quantum on every exchange of messages and gives every voice and video call its own post-quantum key. WhatsApp and Messenger have not announced any post-quantum protection of conversation content so far, and Telegram does not encrypt end-to-end by default at all.
| Szept | Signal | Messenger | iMessage | Telegram | ||
|---|---|---|---|---|---|---|
| End-to-end encryption by default | yes — chats, groups, calls | yes | yes | yes (since 2023) | yes — between Apple devices only | no — secret 1:1 chats only, groups without E2E |
| Post-quantum session setup | yes (ML-KEM-768) | yes (PQXDH, 2023) | not announced | not announced | yes (PQ3, 2024) | no |
| Post-quantum re-keying | yes — on every exchange | yes — SPQR, key split across many messages | not announced | not announced | yes — periodic, not with every message | no |
| Post-quantum key for every call | yes, new for each call | — | not announced | not announced | — | no |
| Who is behind it | independent, Polish; no ads, no tracking | non-profit foundation; no ads | Meta — an advertising company | Meta — an advertising company | Apple | private company; ads in channels |
| Own server for a company | yes | no | no | no | no | no |
| Android and iPhone | yes | yes | yes | yes | Apple only | yes |
| Browser version | yes — Szept Web, end-to-end encrypted in the browser | no — desktop app only | yes — WhatsApp Web | yes — messenger.com | no | yes — Telegram Web |
Swipe the table sideways → As of November 2026, based on vendors' public information. “Not announced” = the vendor has not publicly stated that it protects conversation content post-quantum. “—” = no clear public information.
Frequently asked questions
What is post-quantum encryption?
Encryption built on mathematical problems that neither a regular nor a quantum computer can solve quickly. In Szept it is ML-KEM-768 — the NIST FIPS 203 standard — combined with classical X25519.
Do quantum computers break encryption already?
No. The threat is recording encrypted traffic today and decrypting it in the future. Post-quantum encryption protects the messages you send right now.
Does post-quantum encryption slow Szept down?
Practically not. ML-KEM keys are larger (about 1 KB instead of 32 bytes), so every text message carries about 2.3 KB extra — unnoticeable on today's connections. In calls the exchange happens once, at the start, and does not slow audio or video down.
Do I need to turn anything on?
No. Post-quantum protection works automatically in all conversations, groups and calls.
Do WhatsApp and Messenger have post-quantum encryption?
Both encrypt conversations end-to-end, but as of November 2026 Meta has not announced post-quantum protection of conversation content in either. Without it, traffic recorded today can be decrypted once a large enough quantum computer exists. Szept protects every message and every call against that.
Is Telegram end-to-end encrypted?
Not by default. Regular chats and all groups in Telegram are encrypted only to the server, which can access them. End-to-end encryption exists only in manually started “secret chats” between two people, with no post-quantum protection.
Is Szept more secure than Signal?
In one measurable respect, yes: after a key leak a conversation regains post-quantum protection after a single exchange of messages, while with Signal it takes many. Signal, on the other hand, has a formally verified protocol and years of audits. Szept is a Polish alternative: no ads, post-quantum in messages, re-keying and calls, and deployable on a company's own server.