Signal Ships Automatic Key Verification With Transparency Log

An open handwritten ledger lying on a worn wooden table — illustrative of an append-only log, the structure key transparency uses to record every key change. Photo by Vaan Photography on Pexels

PRIVACY & DATA · 12 Aug 2026 —

MOUNTAIN VIEW, 11 AUG 2026 — Signal has shipped a way to confirm you are talking to the right person without reading a sixty-digit number aloud to them. The feature is called automatic key verification, and it guards against a compromise of Signal itself.

The announcement plainly states the threat model: a malicious party compromises Signal's central directory to associate a different key with your contact's phone number. Everything after that is encrypted correctly — to the wrong person.

What safety numbers were always for

End-to-end encryption guarantees that only the holder of the other key can read your messages. It does not tell you whose key it is. That is the job of the directory, and the directory is the one part of Signal you have to trust.

Safety numbers were the answer: compare a long code with your contact over some other channel, and you have confirmed you hold each other's real keys. It works, and almost nobody does it, because it requires a second channel and a willing counterparty for every contact you care about.

A security control that is correct but unused protects nobody, which is the problem this feature addresses.

How the new one works

The mechanism, key transparency, offers a different kind of guarantee.

Signal records registrations, phone-number changes and username changes in an append-only log tree, searchable through prefix trees. Public identifiers are processed through a verifiable random function, and the values they map to are protected by a keyed hash, so the log is auditable without being a readable directory of who uses Signal.

Cloudflare and Trail of Bits each act as independent auditors, cryptographically signing entries so that everyone is shown one consistent view of the log. Your app also periodically checks your own identifiers in it.

The user-facing part is one tap. Open a contact's profile, choose View Safety Number, press Verify automatically, and a green checkmark reading "Encryption verified" appears if it succeeds.

What it does not do, in Signal's own words

This is the most useful part of the announcement, and it is unusually candid for a security launch.

Situation	Automatic verification
You only know a username	Cannot verify. You need the contact's phone number.
Contact changed number	Fails — your device cannot distinguish that from server interference.
Account fully taken over	Not detected. Key transparency does not verify who controls an account.
Timing	Not guaranteed to catch interference the moment it happens.

That third row is the one to remember. Key transparency proves the association between an identifier and a key is globally consistent — that everyone sees the same key for that phone number. It says nothing about whether the person holding the phone is your friend. If someone takes over the account outright, the key is legitimately theirs and the log will happily agree.

Signal spells this out: additional verification, such as following up after a safety-number change, is still needed to catch a full account compromise.

Why the auditors matter more than the cryptography

Transparency logs are only as good as the guarantee that everyone is being shown the same log. A server that can present one version to you and another to your contact defeats the whole construction — this is the split-view problem, and it is what auditors exist to prevent.

Cloudflare and Trail of Bits signing entries is therefore not a compliance flourish; it is the load-bearing part. It also means the property being relied upon is now partly organisational: Signal, Cloudflare and Trail of Bits would all have to be compromised or coerced together.

That is a meaningfully harder ask than compromising one directory, and it is not the same as a mathematical impossibility. Signal notes the auditors never see plaintext user data, which is right for privacy and does limit what they can independently notice.

What to actually do

If you have contacts whose safety numbers you never verified — which for most people is all of them — open a few profiles and press the button. It costs seconds and closes the gap that manual verification left open for years.

Keep treating a safety-number change notification as an event that needs a human explanation, not a tap-through. That is still the signal that something changed at the other end, and it is the one automatic verification explicitly cannot interpret for you.

And if your threat model includes someone taking the other person's phone, none of this addresses that. Nothing in cryptography does.

What to watch

Whether other messengers adopt key transparency. The construction is not Signal-specific, and the argument for it — that users cannot be expected to verify manually at scale — applies to every encrypted messenger with a central directory.

Whether the auditor set grows. Two independent auditors is materially better than none and a smaller number than the property deserves.

And whether username-only contacts ever get covered. That gap matters most for exactly the people who chose usernames to avoid revealing a phone number, which is to say those with the strongest reason to verify.

Priya Nair’s Verdict Privacy & Digital Trust

Signal has closed a gap that existed because the old control was correct and nobody used it. Safety numbers work perfectly and require a second channel and a willing counterparty for every contact, which in practice meant almost no one verified anyone. A control that is right and unused protects nobody, so replacing sixty digits with one tap is a real security gain even though the cryptography underneath is not stronger. Read the limitations, though, because Signal states them more honestly than most launches do. Key transparency proves the association between an identifier and a key is globally consistent — that everyone sees the same key for that number. It does not prove who is holding the phone. If an account is taken over outright, the key is legitimately theirs and the log agrees. So a safety-number change notification still needs a human explanation rather than a tap-through; that is the one thing this cannot interpret for you. And the load-bearing part is not the maths, it is Cloudflare and Trail of Bits auditing the log so nobody can be shown a private version of it. That is an organisational guarantee wearing a cryptographic coat, which is meaningfully harder to defeat than one directory and is not the same thing as impossible.

Sources & cross-checks

Primary: Signal's own announcement, “Introducing Automatic Key Verification”, 11 August 2026 — for the threat of a malicious party compromising Signal and associating a different key with a connection's phone number; the key-transparency mechanism recording registrations, phone-number changes and username changes in a log tree searchable via prefix trees; public identifiers processed through a verifiable random function with mapped values protected by a keyed hash function; Cloudflare and Trail of Bits each serving as independent auditors cryptographically signing entries to maintain one consistent view; the app periodically checking your own identifiers; the user flow of View Safety Number then “Verify automatically”, showing a green checkmark and “Encryption verified”; and the stated limits — username-only contacts cannot be verified without a phone number, a contact's number change fails because “your device can't distinguish that from server interference”, the system “does not verify the identity of the user who controls a particular phone number or username”, auditors “never see any plaintext user data”, and checks are “neither guaranteed to catch Mallory's interference the moment it happens”.
Corroborated: The Register, 11 August 2026, reporting the same feature as an additional layer for confirming you are talking to the intended person.
What could not be established: ⚠ We could not establish the rollout schedule across Android, iOS and desktop, whether the feature is on by default, how frequently the app re-checks the log, what a verification failure looks like to a user, or whether the auditor set is expected to grow. We could not open secondary coverage beyond The Register, so the detail here rests almost entirely on Signal's own post.
Method and limits: Compiled 11 August 2026. The argument that an unused control protects nobody, the framing of the split-view problem as the reason auditors are load-bearing rather than decorative, the observation that the resulting guarantee is partly organisational, and the advice to keep treating safety-number change notifications as events needing explanation are RECATOOLS analysis rather than claims by Signal. Readers should take instructions from Signal's own post rather than from this article.

Tags: #Cybersecurity #Privacy #Messaging

Priya Nair

Data, AI Governance & Policy Analyst

Priya Nair covers AI governance, data protection, privacy, and digital trust topics for RECATOOLS.

View author profile → · Editorial policy

About this byline Priya Nair is a RECATOOLS editorial persona for AI governance, privacy, and digital trust coverage. Articles are produced and reviewed under RECATOOLS editorial supervision.

Corrections policy