Join our Newsletter — 33% off our NHI Course
Home› Guides› End-to-End Encryption and Messaging Protocols Guide: Signal, MLS…
Guide Security Policies & Awareness

End-to-End Encryption and Messaging Protocols Guide: Signal, MLS and Key Transparency Explained

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 27 October 2026 13 min read
On this page

End-to-end encryption means only the communicating endpoints hold the keys, so the service operator, a network attacker or a compromised server cannot read messages, but it also means identity, key verification and the endpoint itself now carry the security of the conversation. Modern secure messaging rests on a small number of designs: the Signal protocol family (X3DH or PQXDH key agreement and the Double Ratchet), the IETF Messaging Layer Security protocol (MLS, RFC 9420) for groups, and key transparency to stop a server quietly swapping keys. This guide explains what each provides and does not provide, how to evaluate an end-to-end encrypted product, how enterprises reconcile it with retention and investigations, and where the lawful access debate affects risk decisions. It is written for security architects, policy owners and awareness leads who set messaging policy for staff.

Key takeaways

  • E2EE protects content, not metadata or endpoints. Who talks to whom, when and how often is often visible to the service, and malware or a lost unlocked phone reads messages after decryption.
  • Forward secrecy and post-compromise security are the two properties to ask for. The Double Ratchet and MLS both refresh keys so that one stolen key exposes a bounded window, not the whole history or future.
  • Key verification closes the remaining gap. Signal's PQXDH specification states that without comparing identity keys through an authenticated channel, the parties have no cryptographic guarantee of who they are talking to; key transparency automates much of that check.
  • Account recovery is the weakest link. Phone number and SMS-based registration let an attacker who controls a number try to register it on a new device, as Signal users exposed through the 2022 Twilio breach learned.
  • Enterprise policy must choose deliberately. Consumer E2EE apps defeat retention and eDiscovery by design, so regulated firms need approved tools and rules about which conversations may use them.

What end-to-end encryption guarantees

PropertyMeaningProvided by
Confidentiality from the serverThe operator relays ciphertext it cannot decryptAll E2EE protocols, if keys are generated and held on endpoints
AuthenticationYou are talking to the holder of a particular identity keyIdentity keys plus verification (safety numbers, QR codes, key transparency)
Forward secrecyStealing today's keys does not expose past messagesPer-message keys deleted after use (Double Ratchet); epoch keys in MLS
Post-compromise security (break-in recovery)After a compromise ends, future messages become secure againDiffie-Hellman ratchet; MLS updates and commits
Post-quantum forward secrecyRecorded traffic stays secret against a future quantum computerPQXDH, ML-KEM Braid, hybrid suites
DeniabilityA transcript does not cryptographically prove authorship to third partiesX3DH and PQXDH (a form of deniability)
Not providedMetadata privacy, endpoint security, protection from a malicious contact, message retentionOther controls

The Signal protocol family

Signal publishes its specifications openly. They are used in Signal itself and in other products that license or reimplement the design.

  • X3DH: establishes a shared secret between two parties who mutually authenticate with public keys, with forward secrecy and deniability, while the recipient is offline.
  • PQXDH: (revision 3, dated 24 May 2023, last updated 23 January 2024) adds a post-quantum KEM to the key agreement for post-quantum forward secrecy. The KEM is a parameter; the specification's example is CRYSTALS-Kyber-1024. Authentication still relies on discrete log hardness, so it is not quantum-secure in that revision.
  • Prekeys: each user publishes an identity key, a signed prekey rotated periodically, one-time prekeys, and in PQXDH a signed last-resort PQ prekey and signed one-time PQ prekeys. Signatures stop a malicious server substituting forged prekeys, but a server can still withhold one-time prekeys, weakening initial forward secrecy, which is why the specification recommends rate limits on prekey fetches.
  • Double Ratchet: (revision 4, dated 4 November 2025) derives a unique key for every message through a symmetric-key ratchet and mixes fresh Diffie-Hellman output into a root chain each time the parties exchange new ratchet keys. That gives forward secrecy and break-in recovery.
  • Sesame: manages sessions across multiple devices per user, which is where many real-world complexities (linked desktops, new phones) appear.
  • ML-KEM Braid: a sparse continuous key agreement using ML-KEM that adds post-quantum forward secrecy and post-compromise security to the ratchet.

The Double Ratchet specification also documents a practical trade-off defenders should know. Messages that arrive out of order require stored skipped message keys. A malicious sender can force storage growth, and an attacker who recorded the lost messages and later compromises the device could use stored keys, so implementations cap skipped keys (the specification gives 1,000 per session as an example) and delete them after an interval. The specification also assumes deleted data cannot be recovered through low-level device access, which places device forensics outside its protection.

Messaging Layer Security (MLS) for groups

Pairwise protocols scale poorly for large groups because each message must be encrypted separately to each member. MLS (RFC 9420, Standards Track, July 2023) provides asynchronous group key establishment for groups from two to thousands of clients with forward secrecy and post-compromise security.

  • Epochs: a group moves through a linear sequence of epochs. Members of an epoch share an epoch secret from which message and authentication keys are derived; membership can change between epochs.
  • Ratchet tree (TreeKEM): members sit at the leaves of a binary tree and know only the private keys on their path to the root, so updates cost logarithmic rather than linear work.
  • Proposals, commits and welcomes: Add, Update and Remove proposals describe changes; a Commit applies them and adds fresh entropy; a Welcome lets a new member initialise state. RFC 9420 recommends members refresh keys on the order of hours or days.
  • Two services: the Delivery Service routes messages and stores KeyPackages and is treated as largely untrusted. The Authentication Service binds credentials (basic or X.509) to signing keys and is trusted; RFC 9420 section 16.10 describes the impact of its compromise.
QuestionSignal-style pairwise sessionsMLS
Group sizeEfficient for one-to-one and small groupsDesigned for groups up to thousands
StandardisationOpen specifications published by SignalIETF Standards Track (RFC 9420)
Identity bindingIdentity keys verified by users or key transparencyCredentials validated by the Authentication Service, including X.509
Enterprise fitUsually consumer identities (phone numbers, usernames)Credentials can be bound to corporate identity, useful for federated enterprise messaging
Removing a memberStop encrypting to their devicesRemove proposal and commit move the group to a new epoch the removed member cannot read

Key verification and key transparency

Every E2EE system depends on the server handing out the right public key. A server that is compromised or compelled could supply its own key and sit in the middle (ATT&CK T1557 Adversary-in-the-Middle describes the general technique). Three controls address this.

  1. Manual verification: comparing safety numbers, fingerprints or QR codes in person or over another authenticated channel. Strong, but few users do it.
  2. Change notifications: clients warn when a contact's identity key changes. Users must be taught that a change after a new phone is normal, and a change with no explanation is worth a call.
  3. Key transparency: the service publishes key bindings in an append-only, verifiable log, so clients can check that everyone sees the same key for a given identity. The IETF KEYTRANS working group is standardising this: at 27 October 2026, draft-ietf-keytrans-architecture-10 (7 October 2026) was submitted to the IESG for publication as Informational, and draft-ietf-keytrans-protocol-05 (5 July 2026) was a working group draft. Both are drafts and can change.

Apple's iMessage Contact Key Verification is an example of the product side: it automatically alerts both parties when verification fails, supports comparing verification codes and a public verification code, and requires two-factor authentication, iCloud Keychain and recent OS versions on every signed-in device. Apple notes that it is not supported with Managed Apple Accounts and is not designed to prevent phishing or scams. That limitation is the general rule: verification proves which key you are talking to, not that the person behind it is trustworthy.

Where E2EE fails in practice

Attacks on well-designed E2EE rarely break the cryptography. They target the identity and device around it.

Weak pointExampleDefensive control
Account registration and recoverySMS codes intercepted or phone number taken over to register a new deviceRegistration lock or PIN, alerts on new device registration, not using SMS as the only factor
Linked devicesAn attacker links a desktop or web client to a victim account through a phishing QR codePeriodic review of linked devices; user training on device-linking prompts
Endpoint compromiseSpyware reads decrypted messages or notifications on the phone (mobile ATT&CK T1517 Access Notifications)Mobile device management or mobile threat defence, OS updates, lockdown modes for high-risk users
Cloud backupsMessage backups stored with keys the provider controlsEnd-to-end encrypted backups or backups disabled by policy
Wrong participantA journalist or official added to the wrong groupVerified contacts, naming conventions, membership review before sensitive discussions
MetadataContact graphs and timing reveal relationshipsChoose services that minimise metadata retention; treat metadata as sensitive

Enterprise use: retention, eDiscovery and approved tools

For an organisation, E2EE creates a policy tension. It protects staff and customers from interception and provider breaches, but it also removes the organisation's ability to retain, supervise and search business communications if those happen in consumer apps.

  • Classify the use: decide which conversations are business records (instructions, approvals, client advice) and must stay in retained channels, and which personal or security-sensitive conversations may use consumer E2EE apps. See the Enterprise Messaging Retention and Off-Channel Communications Guide.
  • Approved tools: where E2EE is needed for business, choose products that bind identities to the corporate IdP, support device management, and offer an archiving path that is explicit and auditable rather than hidden.
  • Retention and legal hold: align disappearing-message settings with retention policy and legal hold obligations. See the eDiscovery and Legal Hold Guide.
  • Incident response: keep an out-of-band E2EE channel for incident coordination when corporate systems may be compromised, with members verified in advance and documented.
  • Bots and integrations: any bot, archiver or integration that joins an E2EE conversation is a participant that can read everything. Treat it as a non-human identity with an owner, scoped permissions and its own credential lifecycle.

Small organisations usually need one approved consumer app for out-of-band use plus clear retention rules for business chat. Large and regulated organisations need an approved enterprise messaging platform, archiving controls where regulation requires them, and a documented exception path for E2EE consumer apps.

The lawful access debate and risk decisions

Governments in several jurisdictions have argued for law enforcement access to encrypted content, while cryptographers and many providers argue that any exceptional access mechanism weakens security for everyone. Security teams do not need to settle the debate, but they do need to understand that the legal environment can change the guarantees a product gives.

The United Kingdom illustrates the mechanism. Section 253 of the Investigatory Powers Act 2016 lets the Secretary of State give a technical capability notice to a telecommunications or postal operator, subject to necessity, proportionality and approval by a Judicial Commissioner. The obligations that may be imposed include a requirement to remove electronic protection applied by or on behalf of the operator. The Investigatory Powers (Amendment) Act 2024 amended the section, in force 14 October 2024, and legislation.gov.uk showed it up to date to 23 October 2026 with one further change outstanding. Other jurisdictions are debating comparable measures; check the current position for every country where your providers operate.

  • Vendor due diligence: ask where keys are generated and held, whether the provider can add devices or participants silently, and whether key transparency or change notification would reveal it.
  • Data location and jurisdiction: record which legal regimes apply to each messaging provider in your supplier risk register.

This section summarises one law for context and is not legal advice.

Rolling out a messaging security policy

  1. Inventory: find which messaging apps staff use for work, including on personal phones, through surveys, MDM data and network telemetry.
  2. Decide the approved set: one retained enterprise platform, one approved E2EE app for defined uses, and a list of prohibited apps for business records.
  3. Configure: enforce registration lock or PIN, disable unencrypted cloud backup where possible, set disappearing-message defaults that match retention policy, and enrol corporate devices in MDM.
  4. Train: cover key change warnings, verification of high-risk contacts, linked-device reviews and the risk of adding the wrong person to a group. See the Securing Messaging Apps Guide.
  5. Pilot and measure: start with executives, incident responders and finance approvers, then extend.
  6. Review: re-check settings after major app updates, because defaults for backups, linked devices and post-quantum features change.

Detection and measurement

  • Alert on new device registrations and linked devices for high-risk users where the platform exposes them.
  • Monitor SIM swap and number port events for executives' numbers through the carrier where available. See the SIM Swap Fraud Guide.
  • Track off-channel communication findings from supervision or audit.
MetricTarget
High-risk users with registration lock or PIN enabled100%
Incident response channel members verified in the last 12 months100%
Business communications found in unapproved appsFalling each review period
Corporate devices with current OS and messaging app versionsAbove agreed threshold
Integrations and bots in approved messaging with a named owner100%

Common failure patterns

  • Trusting the padlock: treating "end-to-end encrypted" as proof of a secure conversation while ignoring endpoints, backups and group membership.
  • SMS as the root of identity: phone number registration without a PIN. In the Twilio breach of 2022, Signal said attackers could have seen about 1,900 users' phone numbers or SMS verification codes and could have tried to re-register those numbers on another device.
  • Ignoring key change warnings: users trained to click through every safety number change.
  • Retention by accident: disappearing messages on by default for business decisions that must be recorded, or off by default for sensitive personal data that should not persist.
  • Unowned bots: archivers or assistants added to encrypted channels with broad access and no owner.

Framework mapping

FrameworkReferenceRelevance
NIST SP 800-53 Rev. 5SC-8, SC-8(1), SC-12, SC-13Protected transmission, key management and approved cryptography
NIST SP 800-53 Rev. 5IA-2, AU-11User authentication to messaging; retention of required records
NIST CSF 2.0PR.DS-02, PR.AA-03, GV.OC-03Data in transit, authentication, legal and regulatory requirements
MITRE ATT&CK v20T1557 (enterprise), T1517 and T1636.004 (mobile)Interception, notification and SMS collection on devices

Practitioner checklist

  • Define which communications are business records and which tools may carry them.
  • Approve one E2EE app for out-of-band and sensitive use, with registration lock, backup and disappearing-message settings documented.
  • Prefer products with key transparency or automatic key verification, and train users to act on key change warnings.
  • Pre-verify members of incident response and executive channels.
  • Review linked devices and group memberships for high-risk users regularly.
  • Treat bots and archivers in encrypted channels as owned non-human identities.
  • Record the jurisdictions and lawful access regimes that apply to each messaging provider.
  • Track post-quantum support (PQXDH, ML-KEM Braid, MLS cipher suites) in approved tools.

Standards and references

Related NHI Mgmt Group resources: Securing Messaging Apps Guide · OpenPGP and S/MIME Email Encryption Guide · Hybrid Post-Quantum TLS Key Exchange Guide · Enterprise Messaging Retention and Off-Channel Communications Guide

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 27 October 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.