The forgot password flow is an authentication path that skips the password by design, so any weakness in it, from a guessable token to a reset link built from a spoofed Host header, is a direct route to account takeover. Reset flows also leak which accounts exist, can be abused to flood users with messages, and too often let an attacker bypass MFA or keep a session alive after the real owner recovers the account. This guide sets out how to build a reset flow that holds up: request handling without enumeration, token generation and storage, delivery channels, the reset step itself, MFA and recovery codes, help desk resets, and how to detect abuse. It draws on the OWASP Forgot Password Cheat Sheet, OWASP ASVS 5.0 and NIST SP 800-63B-4, and is written for developers, identity engineers and application security reviewers.
Key takeaways
- Same answer, same time, every time. Return an identical message and response time whether or not the account exists, as both the OWASP cheat sheet and ASVS V6.3.8 require.
- Tokens are bearer credentials. Generate them with a cryptographically secure random generator, make them long, store only a hash, bind them to one user, use them once and expire them quickly.
- Never build the link from the request. Use a configured canonical URL, not the Host header, and send the link only to the contact already on record.
- A reset must not bypass MFA. ASVS V6.4.3 requires that resetting a forgotten password does not bypass enabled MFA; recovering a lost factor needs identity evidence at the original enrolment level (V6.4.4).
- Finish the job after the reset. Notify the user, end other sessions, invalidate other outstanding tokens, and do not log the user in automatically.
The two halves of the flow
The OWASP Forgot Password Cheat Sheet splits the service into the request step (the user submits a username or email) and the reset step (the user presents a token or code and sets a new password). Each half has its own failure modes.
| Step | Control | Source |
|---|---|---|
| Request | Consistent message and consistent response time for existing and non-existing accounts; avoid quick exits, for example by sending asynchronously | OWASP cheat sheet; ASVS V6.3.8 |
| Request | Per-account rate limiting, CAPTCHA or similar to stop floods of reset messages | OWASP cheat sheet |
| Request | Do not change the account (for example lock it) until a valid token is presented | OWASP cheat sheet |
| Delivery | Send only to the address or number on record; build URLs from configuration over HTTPS | OWASP cheat sheet |
| Reset page | Referrer-Policy: no-referrer so the token does not leak to third-party resources; rate-limit token attempts | OWASP cheat sheet |
| Reset | Apply the normal password policy; store with the normal password hashing scheme | OWASP cheat sheet |
| After reset | Notify the user (never include the password); require normal login; offer or force termination of other sessions | OWASP cheat sheet; ASVS V6.3.7, V7.4.3 |
Handling the request without leaking accounts
- Message: use wording such as "If an account exists for that address, we have sent instructions." Show it for every submission, valid or not.
- Timing: the slow path (token generation, database write, email queueing) must not run only for real accounts. Put work on a queue and return immediately for all requests, or perform equivalent work for unknown addresses.
- Other channels: registration and "change email" forms leak the same information if they say "address already in use". ASVS V6.3.8 explicitly includes registration and forgot password functionality.
- Abuse limits: cap reset requests per account and per source, and use bot controls on the form. Floods are both a harassment tool and a way to hide a real reset email among many. See the Rate Limiting and API Abuse Guide and the Bot Management Guide.
- No lockout: the OWASP cheat sheet advises that accounts should not be locked in response to forgot password attacks, because that lets anyone deny access to known usernames.
Tokens and codes: generation, storage and lifetime
OWASP's general rules for reset identifiers are that they are generated with a cryptographically secure random number generator, long enough to resist brute force, linked to an individual user in the database, invalidated after use and stored securely. NIST SP 800-63B-4 gives concrete numbers for recovery codes in its account recovery section:
| Item | Requirement or practice | Source |
|---|---|---|
| Saved recovery codes (issued at enrolment, kept by the user) | At least 64 bits from an approved random bit generator; stored hashed; each invalidated after one use | NIST SP 800-63B-4 |
| Issued recovery codes (sent at recovery time) | At least six decimal digits or equivalent from an approved random bit generator | NIST SP 800-63B-4 |
| Validity of issued codes | At most 10 minutes for text or voice, 24 hours for email, 21 days for contiguous US postal addresses and 30 days for other postal addresses | NIST SP 800-63B-4 |
| URL tokens in email links | Long random value (for example 128 bits or more), hash stored server-side, single use, short expiry | OWASP cheat sheet (principles); values are common practice |
| SMS PINs | 6 to 12 digits, creating a limited session that only permits a password reset | OWASP cheat sheet |
| Signed tokens such as JWTs | Possible, but OWASP notes they can introduce additional vulnerabilities; a random token with server-side state is simpler to revoke | OWASP cheat sheet |
Implementation details that matter in review:
- Store a hash, compare in constant time: a database leak of plaintext reset tokens is an instant mass account takeover.
- One live token per account: issuing a new token invalidates the previous one; a successful reset invalidates all outstanding tokens and codes.
- Bind the token to state: include the account ID and a password or credential version, so a token minted before a password change cannot be used after it.
- Consume atomically: mark the token used in the same transaction that changes the password, so two parallel submissions cannot both succeed. See the Race Condition Vulnerabilities Guide.
- Beware link scanners: email security gateways fetch links. Make the token landing page show a form that requires a user action (POST) before consuming the token, so a scanner's GET does not burn or use it.
Delivery: where the link goes and how it is built
The two most damaging reset flaws are both about delivery.
- Host header poisoning: if the application builds the reset URL from the request's Host or
X-Forwarded-Hostheader, an attacker requests a reset for the victim with their own domain in the header, and the victim receives a genuine email whose link sends the token to the attacker. The OWASP cheat sheet says the URL should be hard-coded or validated against trusted domains. See the Host Header Attacks and Validation Guide. - Requester-supplied destination: any flow that lets the requester choose or add the address that receives the reset is broken by design. The Meta AI Instagram account takeover is a recent example: an AI support assistant linked an attacker-supplied email to target accounts and sent the reset code there, and accounts with two-factor authentication enabled were not taken over.
Further rules: send links over HTTPS only; avoid putting the account email in the URL; and, for recently changed contact details, apply a cooling-off period before they can receive resets. The OWASP cheat sheet also warns against relying solely on recently added or changed recovery information when recovering a possibly compromised account. Email-based reset is effectively a magic link: see the Magic Links and Email OTP Security Guide.
MFA, recovery codes and assurance levels
A password reset is only safe if the rest of the authentication stack still applies afterwards.
- Password reset with MFA enrolled: after the token is validated, require the user's second factor before the new password takes effect, or require it at the next login. ASVS V6.4.3 prohibits a reset process that bypasses enabled MFA.
- Lost second factor: ASVS V6.4.4 requires identity proofing evidence at the same level as enrolment. Saved recovery codes issued at enrolment are the usual self-service answer.
- NIST recovery rules: SP 800-63B-4 recognises four recovery method classes (saved recovery codes, issued recovery codes, recovery contacts and repeated identity proofing). For accounts at AAL2, and AAL3 with IAL1 or IAL2, recovery requires two codes from different methods, or one code plus a bound single-factor authenticator, or repeated identity proofing. NIST also notes that replacing a forgotten password while the user can still authenticate with another authenticator is treated as binding a new authenticator, not account recovery.
- No security questions: ASVS V6.4.2 requires that password hints and knowledge-based "secret questions" are not present. The OWASP cheat sheet says they must never be the sole mechanism. New designs should leave them out.
For assurance levels in more depth, see the NIST SP 800-63B-4 Authenticator Assurance Levels Guide. Moving users to passkeys reduces how often reset is needed at all; see the Passwordless and Passkeys Guide.
Help desk and administrator resets
Self-service is not the only reset path. Help desk and administrator resets are a favourite target for social engineering.
- Administrators initiate, users choose: ASVS V6.4.6 says administrators can start a reset for a user but must not be able to set or choose the user's password.
- Verify with something the caller cannot be coached into: call-back to a number on record, video verification against an ID, or a push to an existing enrolled device, never knowledge questions alone.
- Self-service reset is not proof of identity on its own: in the Storm-2949 Azure attack, attackers started self-service password reset and, posing as IT support, persuaded users to approve the MFA prompts; they then removed the users' authentication methods and enrolled their own authenticator. Treat an approved MFA prompt as weak evidence when the request was not initiated by the user, use number matching or phishing-resistant factors, and alert when methods are removed and replaced right after a reset.
See the Account Recovery and Help Desk Security Guide for help desk procedures.
Completing the reset safely
- Validate the token: hash it, look it up, check expiry, user binding and credential version, and rate-limit attempts.
- Enforce MFA: if the account has MFA, require it now.
- Set the password: enter twice, check the normal policy including breached password screening, hash with the normal scheme. See the Password Hashing and Key Derivation Guide.
- Consume and clean up: mark the token used and invalidate every other outstanding reset token and recovery code that was issued but not used.
- End sessions: terminate other sessions and revoke refresh tokens, or at least offer to, as ASVS V7.4.3 requires the option after any authentication factor change.
- Notify: send a notification to the contact details on record (NIST requires notification after account recovery to all notification addresses except postal ones, with instructions for reporting an unauthorised change).
- Return to login: do not log the user in automatically; OWASP notes that auto-login adds complexity to session handling and increases the chance of vulnerabilities.
When the reset follows a suspected compromise, the OWASP cheat sheet adds that you should review recovery addresses, phone numbers and MFA methods with the verified owner, remove unauthorised changes, and not automatically restore superseded recovery details. See the Account Takeover Response Playbook.
Testing and rollout
- Automated tests: identical responses and timings for known and unknown accounts; foreign Host and
X-Forwarded-Hostheaders produce no change in emailed links; tokens are single-use under parallel submission; old tokens fail after a new one is issued or after a password change; MFA is required after reset; sessions end. - Manual review: trace every path that can change a password or contact detail (self-service, support tooling, admin consoles, chatbots and AI assistants, mobile apps, APIs) and confirm each uses the same token service and checks.
- Rollout: when changing token formats, accept old tokens until their natural expiry, then remove the old code path entirely. Roll back by redeploying the previous service version, not by relaxing expiry or single-use checks.
Decision guidance
- Consumer applications: email link plus mandatory MFA where enrolled, saved recovery codes for lost factors, and a slower, evidence-based support route for the rest.
- Workforce identity: self-service reset only with strong factors (passkeys or number-matched push), Temporary Access Pass or equivalent issued by the help desk after verification, and alerts on method changes.
- High-assurance services: follow NIST AAL2 or AAL3 recovery requirements, which rule out single-factor recovery for identity-proofed accounts.
Detection and measurement
Log each reset request (account, source, user agent, Host and forwarded headers), token issuance, token validation failures, password change, sessions ended, and any contact or MFA method change. Alert on:
- Spikes of reset requests for many accounts from few sources (enumeration or spraying) or for one account (harassment or cover).
- Reset followed quickly by MFA method removal or new device registration (ATT&CK T1098.005 Device Registration).
- Token validation from a different country or ASN than the request, which can indicate relayed or poisoned links.
- Requests with non-canonical Host values on the reset endpoint.
| Metric | Target |
|---|---|
| Reset paths using the central token service | 100% (including support tooling and assistants) |
| Response time difference between known and unknown accounts | Not measurable above noise |
| Account takeovers whose entry point was reset or recovery | Tracked and trending down |
| Resets followed by MFA method change within one hour | Each one reviewed |
| Median token lifetime actually configured | Within policy (minutes for SMS, hours at most for email) |
Common failure patterns
- Distinct responses: "No account found" messages or measurable timing differences enable enumeration.
- Request-derived links: reset URLs built from the Host header.
- Requester-chosen destination: support staff, chatbots or AI assistants sending codes to an address the requester supplies, as in the Meta AI case.
- Reusable or long-lived tokens: tokens that survive use, survive a password change, or last for days.
- MFA bypass: the reset logs the user straight in, skipping the second factor.
- Sessions left alive: the attacker's existing session continues after the owner resets the password.
Framework mapping
| Framework | Reference | Relevance |
|---|---|---|
| OWASP ASVS 5.0 | V6.3.7, V6.3.8, V6.4.1 to V6.4.4, V6.4.6, V7.4.3, V7.5.1 | Notification, anti-enumeration, secure codes, no secret questions, no MFA bypass, admin resets, session termination, re-authentication for sensitive changes |
| NIST SP 800-63B-4 | Account recovery and notification sections | Recovery methods, code strength and validity, notification requirements |
| NIST SP 800-53 Rev. 5 | IA-5 Authenticator Management; IA-11 Re-authentication; AC-7 Unsuccessful Logon Attempts | Reset token handling, re-authentication, attempt limits |
| NIST CSF 2.0 | PR.AA-01 | Identity and credential management |
| MITRE ATT&CK | T1621 Multi-Factor Authentication Request Generation; T1098.005 Device Registration; T1110 Brute Force | Prompt bombing during resets, persistence after takeover, token guessing |
| MITRE CAPEC | CAPEC-50 Password Recovery Exploitation | The attack pattern this guide defends against |
For how attackers subvert authentication processes more broadly, see the MITRE ATT&CK T1556 Modify Authentication Process Guide.
Practitioner checklist
- Return the same message and the same response time for every reset request, and apply the same rule to registration and email change forms.
- Rate-limit reset requests per account and per source; never lock accounts because of reset requests.
- Generate tokens with a CSPRNG, store only hashes, bind them to the user and credential version, and expire them quickly.
- Build reset links from a configured canonical URL and send them only to the contact on record.
- Set
Referrer-Policy: no-referreron the reset page and require a user action before consuming the token. - Require MFA after reset when it is enrolled; use saved recovery codes or proofing for lost factors.
- Remove security questions and password hints.
- Consume tokens atomically, invalidate other outstanding tokens, end other sessions and notify the user.
- Route every reset path, including help desk tools and AI assistants, through the same token service.
- Alert on resets followed by MFA method changes or new device registrations.
Standards and references
- OWASP: Forgot Password Cheat Sheet
- OWASP: Application Security Verification Standard 5.0 (V6 Authentication, V7 Session Management)
- NIST: SP 800-63B-4 Authenticator Event Management (account recovery)
- OWASP: WSTG Testing for Host Header Injection
- MDN: Referrer-Policy header
- MITRE: CAPEC-50 Password Recovery Exploitation
Related NHI Mgmt Group resources: Host Header Attacks and Validation Guide · Account Takeover Prevention Guide · Account Recovery and Help Desk Security Guide · Password Security Guide · MITRE ATT&CK T1556 Modify Authentication Process Guide