Skip to content

Privacy-first password manager with built-in email aliasing. Fully encrypted and self-hostable.

12.6 KB
Raw Blame

Security Policy

Important: please report any potential security issues to the email below rather than opening public issues.

We appreciate the security community and responsible researchers who report issues: your findings help us improve AliasVault for everyone. We investigate every report and work with researchers on coordinated disclosure. Vulnerabilities that breach the security boundary (see Section 2.1) may qualify for a CVE identifier; others may still result in fixes or defense-in-depth improvements.

Supported versions

During the public beta we only support the latest major and minor release. On the current 0.x line, that means the latest minor only. Security fixes are only applied to the latest supported version: make sure to upgrade frequently to stay supported.

1. Reporting a Vulnerability

Email: [email protected]

Please include:

  1. Description of the issue
  2. Steps to reproduce
  3. Affected component(s)
  4. Security impact (what an attacker could achieve)
  5. Prerequisites for exploitation
  6. Suggested remediation (optional)

Before reporting, please check the open GitHub issues and the latest main branch: issues already tracked or fixed there count as duplicates (see Section 4.1).

We acknowledge reports within 48 hours. Please do not publicly disclose issues before coordinated resolution. We will credit reporters where applicable (with permission) when fixes are released.


2. Threat Model

AliasVault is a zero-knowledge, end-to-end encrypted password and email alias manager. The full threat model (assets, trust assumptions, adversaries, and what the server knows and can do) is published in the Security Whitepaper. This section is the condensed version that the vulnerability classification below refers to; where the two differ, the whitepaper is authoritative.

Note: We only request CVEs for vulnerabilities that breach the encryption/access control boundary (defined in Section 2.1). Post-compromise scenarios (issues that assume an attacker has already compromised the user's device, runtime environment, or account through external means) are not considered CVE-worthy, though we may still address them as defense-in-depth improvements.

2.1 Primary Security Boundary

The primary security boundary (also referred to as the encryption/access control boundary) in AliasVault is the end-to-end encryption and access control layer that protects user secrets. Concretely, this boundary guarantees that only authorized users can decrypt or access their vault secrets.

  • Vault data and stored email contents are encrypted client-side (email: on the server at the moment of receipt) with keys that only the clients hold; the master password unlocks them and never leaves the device.
  • The server never has access to plaintext vault data or to any key that decrypts it.
  • Network interception, server compromise, or API token leakage must not expose decrypted secrets.

A vulnerability is considered a Class 1 security boundary failure if it allows crossing this boundary without the user's informed and specific authorization of the relevant device, session, or security-sensitive operation.


2.2 Trust Assumptions

The boundary above holds under the following assumptions.

  • Trusted device and operating system. The device on which a vault is unlocked is not compromised: no malware with the user's privileges, no hostile root or jailbreak tooling, and the OS keychain, keystore, process isolation and screen lock work as documented. Everything AliasVault decrypts exists on that device, so a compromised device sees what the user sees.
  • Authentic client code. The user runs an unmodified AliasVault client from an official channel. The browser extensions and mobile apps are installed once (extension and app stores, F-Droid, GitHub releases) and load no code from the server. The web app is delivered by the server on every load, so for the web app the server operator and the TLS connection are trusted to deliver the published code.
  • A strong, secret master password. Argon2id makes guessing expensive, not impossible; anyone holding a copy of the encrypted data can run an offline guessing attack, and no second secret is mixed in.
  • Sound standard cryptography. Argon2id, AES-256-GCM, RSA-OAEP, Ed25519, SRP-6a, HKDF and HMAC with SHA-256, and the platform random number generator.
  • Shared vault members are trusted with that vault. Removal stops future access through the server; it cannot take back what a member already synced.
  • TLS protects the transport, but vault confidentiality does not depend on it. A broken TLS connection exposes session tokens and metadata, not vault contents.
  • The server operator is trusted at the moment an email arrives. Incoming email reaches the SMTP service in plaintext and is encrypted in memory before storage; email encryption protects stored email, not email as it arrives.

2.3 Threat Actors

  • A malicious or compromised server, including the operator, an insider, an intruder, or a leaked database or backup: it may read and modify everything it stores and control every API response.
  • A network attacker between client and server.
  • A malicious website visited while the browser extension is installed.
  • Another AliasVault user, including a current or former member of a shared vault.
  • An attacker holding a stolen API session token.
  • An attacker holding a copy of the encrypted data (server database, device storage, backup or export file).
  • A sender of crafted email to an alias.

None of these may obtain decrypted vault or email contents, learn the master password or any key, make a client accept forged or weakened data, or gain access to a vault they were not granted. The whitepaper lists the mechanism for each adversary.


2.4 Out-of-Scope: Compromised Devices and User Choices

The following represent device compromise scenarios and are outside AliasVault's primary security boundary:

  • Rooted or jailbroken devices
  • Kernel-level or runtime instrumentation (Frida, Xposed, debuggers)
  • Malicious accessibility services
  • Memory dumping of a running process
  • Forensic extraction of device storage or backups
  • Malware operating with equivalent privileges to the user

AliasVault implements defense-in-depth protections in these areas, but issues that require these conditions are classified as local hardening, not core security boundary failures.

Also outside the boundary: a weak, reused or disclosed master password; social engineering where the application accurately presents the security-sensitive action and the user approves it (see Section 3.2.2); tampered client builds and compromise of the channels that deliver them; availability (a malicious server can always refuse to serve or delete data); and the limitations listed in the whitepaper, which are known and tracked and do not need to be reported again.


3. Vulnerability Classification

This section defines how reported issues are categorized.


3.1 Class 1 Vulnerabilities (Project CVE Candidates)

Class 1 issues are considered eligible for CVE assignment under AliasVault's vulnerability disclosure policy. These represent vulnerabilities that breach AliasVault's encryption/access control boundary (see Section 2.1).

A report must:

  • Affect a released version on the supported release line
  • Be reproducible
  • Have a concrete security impact
  • Not depend on full device compromise

3.1.1 Cryptographic Failures

  • Recovery of plaintext vault data without the master password
  • Key derivation or encryption flaws weakening confidentiality
  • SRP (Secure Remote Password) implementation flaws enabling credential recovery or offline password attacks

3.1.2 Authentication & Access Failures

  • Access to another user’s vault or metadata
  • Privilege escalation
  • Authentication bypass
  • Flaws in client or session validation in authentication flows (e.g., passkey or token verification bypass)

3.1.3 Remote Exploitation

  • Remote code execution from malicious input
  • Injection leading to cross-user data access (e.g., XSS, SQL injection, or similar attacks that result in cross-tenant data exposure)
  • Server-side request forgery (SSRF) reaching internal or restricted systems

3.1.4 Secret Exposure

  • Decrypted secrets exposed via logs, APIs, or unintended channels
  • Server-side plaintext exposure
  • Backend returning sensitive internal resources to an attacker

3.2 Security Hardening Issues (Class 2)

These are security improvements but not typically CVE-eligible. They may still be addressed in future updates as defense-in-depth improvements, without a CVE or security advisory.

3.2.1 Compromised Device Scenarios

Issues that only arise or can only be exploited after an attacker has fully compromised the user's device or runtime environment. For example, issues requiring:

  • Root/jailbreak
  • Runtime instrumentation
  • Memory inspection of a live process
  • OS-level data extraction

3.2.2 Social Engineering

Social engineering or phishing attacks where the application accurately presents the security-sensitive action and its material consequences, and the user intentionally approves that action, are out of scope for CVE assignment.

This exclusion does not apply where an application flaw causes the approval UI to misrepresent, obscure, substitute, or fail to bind the action being approved to the action actually performed.

3.2.3 Local Data Handling

  • Clipboard clearing
  • Screen recording protections
  • Autofill edge cases
  • UI redaction improvements

3.2.4 Metadata & Session Observations

  • Token leakage that does not allow vault decryption
  • Metadata visibility inherent to system design

3.2.5 Theoretical or Non-Reproducible

  • Speculative attack chains
  • No working proof of concept
  • Attacks requiring unrealistic assumptions

3.3 Out-of-Scope Reports (Class 3)

These are not treated as security vulnerabilities.

3.3.1 Non-Security Bugs

  • UI inconsistencies
  • Logic errors without security impact

3.3.2 External or Environmental

  • Social engineering
  • User misconfiguration
  • Third-party service compromise outside AliasVault control

3.3.3 Low-Impact Web Findings

  • Missing HTTP headers without exploit path
  • Rate-limit suggestions
  • Username enumeration through registration or the username availability check on a server with public registration enabled
  • Denial-of-service without data compromise

4. How CVE Decisions Are Made

We will request or pursue a CVE ID only if an issue meets all of the following criteria:

  1. The issue impacts confidentiality, integrity, or authentication
  2. It crosses the encryption/access control boundary without the user's informed and specific authorization of the relevant device, session, or security-sensitive operation (as defined in Section 2.1)
  3. It affects a released version
  4. It is reproducible
  5. It does not rely on full device compromise

CVE eligibility also requires that the issue does not fall within an out-of-scope category defined in Sections 3.3 or 5.

Issues that do not meet these criteria may still result in fixes, announced as regular updates rather than security advisories and tracked as hardening or quality improvements.

4.1 Duplicates

A report is a duplicate when the issue was already reported, tracked publicly, found in our own review, or fixed on main (released or not) before we received it. Duplicates are not credited. If the issue qualifies for a CVE, we request it ourselves, credited to whoever found it first.


5. Scope

In scope:

  • AliasVault server (API, Admin, SMTP)
  • Web client
  • Browser extensions (Chrome, Firefox, Edge, Safari)
  • Mobile apps (iOS, Android)
  • Core cryptographic libraries

Out of scope for CVE assignment by this project:

  • Third-party dependency CVEs: the upstream maintainer or a central authority is responsible for issuing CVEs for their own code; AliasVault will promptly upgrade dependencies or apply patches when such issues arise
  • Device OS vulnerabilities
  • Hardware compromise
  • Phishing, social engineering, or other attacks that rely solely on deceiving the user rather than exploiting the application, except where an application flaw misrepresents, obscures, substitutes, or fails to bind the security-sensitive action being authorized, as described in Section 3.2.2

6. Disclosure Timeline

Stage Target
Acknowledgment 48 hours
Initial assessment 7 days
Fix & release goal 30 days

These are target timelines; complex issues may require adjustment in coordination with the reporter. We follow coordinated disclosure.