STRIDE Threat Modeling

Thinking Tools4.3 Crafting Solutions for the Future StateAlso known as STRIDE

Introduced at Microsoft by Loren Kohnfelder and Praerit Garg (The Threats to Our Products, 1999), described by Michael Howard and David LeBlanc (Writing Secure Code, second edition), and set out in full by Adam Shostack (Threat Modeling, 2014).

Checks each part of a system against six threat types, spoofing, tampering, repudiation, information disclosure, denial of service, and elevation of privilege, and produces a ranked threat list.

Eight threats to a login service, sorted into the six STRIDE types

The top row holds the threats to who someone is and what they may do; the bottom row, threats to the data and the service. The first threat is placed as an example; place the rest by the property each one breaks. A type with nothing in it is a type nobody has checked yet. Illustrative.

Items (select one, then a box)
  • Not placedPlacedSpoofingRepudiationElevation of privilegeTamperingInformation disclosureDenial of service
  • Not placedPlacedSpoofingRepudiationElevation of privilegeTamperingInformation disclosureDenial of service
  • Not placedPlacedSpoofingRepudiationElevation of privilegeTamperingInformation disclosureDenial of service
  • Not placedPlacedSpoofingRepudiationElevation of privilegeTamperingInformation disclosureDenial of service
  • Not placedPlacedSpoofingRepudiationElevation of privilegeTamperingInformation disclosureDenial of service
  • Not placedPlacedSpoofingRepudiationElevation of privilegeTamperingInformation disclosureDenial of service
  • Not placedPlacedSpoofingRepudiationElevation of privilegeTamperingInformation disclosureDenial of service
  • Not placedPlacedSpoofingRepudiationElevation of privilegeTamperingInformation disclosureDenial of service

What is attacked

Identity, rightsData and service
Spoofing
Repudiation
Elevation of privilege
Tampering
Information disclosure
Denial of service
Fakes or altersDenies or leaksSeizes or blocks

What the attacker does

Stolen session cookie. Browser. A cookie copied from a shared computer logs the attacker in as the user.Not placed yet.Stolen session cookie. Browser. A cookie copied from a shared computer logs the attacker in as the user.Spoofing: Pretending to be someone else. Breaks authentication; the usual fix is stronger proof of identity, such as a second factor or short-lived sessions.High. The attacker passes as the user. Fix: bind sessions to the device and expire them after an hour idle.Stolen session cookie. Browser. A cookie copied from a shared computer logs the attacker in as the user.Repudiation: Doing something and denying it later, with no record to prove otherwise. Breaks accountability; the fix is a log the actor cannot edit.Stolen session cookie. Browser. A cookie copied from a shared computer logs the attacker in as the user.Elevation of privilege: Getting rights the user was never given. Breaks authorization; the fix is checking rights on every request, on the server.Stolen session cookie. Browser. A cookie copied from a shared computer logs the attacker in as the user.Tampering: Changing data or code without permission. Breaks integrity; the fix is signing or checking what can be changed.Stolen session cookie. Browser. A cookie copied from a shared computer logs the attacker in as the user.Information disclosure: Exposing data to someone not meant to see it. Breaks confidentiality; the fix is encryption and saying less.Stolen session cookie. Browser. A cookie copied from a shared computer logs the attacker in as the user.Denial of service: Making the service unusable for the people entitled to it. Breaks availability; the fix is limits, quotas, and capacity.Edited login token. Login API. The token's contents are encoded but not signed, so a user can change them.Not placed yet.Edited login token. Login API. The token's contents are encoded but not signed, so a user can change them.Spoofing: Pretending to be someone else. Breaks authentication; the usual fix is stronger proof of identity, such as a second factor or short-lived sessions.Edited login token. Login API. The token's contents are encoded but not signed, so a user can change them.Repudiation: Doing something and denying it later, with no record to prove otherwise. Breaks accountability; the fix is a log the actor cannot edit.Edited login token. Login API. The token's contents are encoded but not signed, so a user can change them.Elevation of privilege: Getting rights the user was never given. Breaks authorization; the fix is checking rights on every request, on the server.Edited login token. Login API. The token's contents are encoded but not signed, so a user can change them.Tampering: Changing data or code without permission. Breaks integrity; the fix is signing or checking what can be changed.High. Anything in the token can be rewritten. Fix: sign tokens and reject any that fail the check.Edited login token. Login API. The token's contents are encoded but not signed, so a user can change them.Information disclosure: Exposing data to someone not meant to see it. Breaks confidentiality; the fix is encryption and saying less.Edited login token. Login API. The token's contents are encoded but not signed, so a user can change them.Denial of service: Making the service unusable for the people entitled to it. Breaks availability; the fix is limits, quotas, and capacity.Admin denies a deletion. Admin console. Admins can delete accounts, and nothing records who did it.Not placed yet.Admin denies a deletion. Admin console. Admins can delete accounts, and nothing records who did it.Spoofing: Pretending to be someone else. Breaks authentication; the usual fix is stronger proof of identity, such as a second factor or short-lived sessions.Admin denies a deletion. Admin console. Admins can delete accounts, and nothing records who did it.Repudiation: Doing something and denying it later, with no record to prove otherwise. Breaks accountability; the fix is a log the actor cannot edit.Medium. A deletion cannot be traced to anyone. Fix: an append-only audit log of admin actions.Admin denies a deletion. Admin console. Admins can delete accounts, and nothing records who did it.Elevation of privilege: Getting rights the user was never given. Breaks authorization; the fix is checking rights on every request, on the server.Admin denies a deletion. Admin console. Admins can delete accounts, and nothing records who did it.Tampering: Changing data or code without permission. Breaks integrity; the fix is signing or checking what can be changed.Admin denies a deletion. Admin console. Admins can delete accounts, and nothing records who did it.Information disclosure: Exposing data to someone not meant to see it. Breaks confidentiality; the fix is encryption and saying less.Admin denies a deletion. Admin console. Admins can delete accounts, and nothing records who did it.Denial of service: Making the service unusable for the people entitled to it. Breaks availability; the fix is limits, quotas, and capacity.Error reveals the email. Login API. A wrong password says incorrect password; an unknown email says no such account.Not placed yet.Error reveals the email. Login API. A wrong password says incorrect password; an unknown email says no such account.Spoofing: Pretending to be someone else. Breaks authentication; the usual fix is stronger proof of identity, such as a second factor or short-lived sessions.Error reveals the email. Login API. A wrong password says incorrect password; an unknown email says no such account.Repudiation: Doing something and denying it later, with no record to prove otherwise. Breaks accountability; the fix is a log the actor cannot edit.Error reveals the email. Login API. A wrong password says incorrect password; an unknown email says no such account.Elevation of privilege: Getting rights the user was never given. Breaks authorization; the fix is checking rights on every request, on the server.Error reveals the email. Login API. A wrong password says incorrect password; an unknown email says no such account.Tampering: Changing data or code without permission. Breaks integrity; the fix is signing or checking what can be changed.Error reveals the email. Login API. A wrong password says incorrect password; an unknown email says no such account.Information disclosure: Exposing data to someone not meant to see it. Breaks confidentiality; the fix is encryption and saying less.Medium. Attackers learn which emails have accounts. Fix: one message for both cases.Error reveals the email. Login API. A wrong password says incorrect password; an unknown email says no such account.Denial of service: Making the service unusable for the people entitled to it. Breaks availability; the fix is limits, quotas, and capacity.Unencrypted backups. Database. Nightly backups of the user table sit unencrypted in shared storage.Not placed yet.Unencrypted backups. Database. Nightly backups of the user table sit unencrypted in shared storage.Spoofing: Pretending to be someone else. Breaks authentication; the usual fix is stronger proof of identity, such as a second factor or short-lived sessions.Unencrypted backups. Database. Nightly backups of the user table sit unencrypted in shared storage.Repudiation: Doing something and denying it later, with no record to prove otherwise. Breaks accountability; the fix is a log the actor cannot edit.Unencrypted backups. Database. Nightly backups of the user table sit unencrypted in shared storage.Elevation of privilege: Getting rights the user was never given. Breaks authorization; the fix is checking rights on every request, on the server.Unencrypted backups. Database. Nightly backups of the user table sit unencrypted in shared storage.Tampering: Changing data or code without permission. Breaks integrity; the fix is signing or checking what can be changed.Unencrypted backups. Database. Nightly backups of the user table sit unencrypted in shared storage.Information disclosure: Exposing data to someone not meant to see it. Breaks confidentiality; the fix is encryption and saying less.High. One leaked file exposes every password hash. Fix: encrypt backups and restrict who can read them.Unencrypted backups. Database. Nightly backups of the user table sit unencrypted in shared storage.Denial of service: Making the service unusable for the people entitled to it. Breaks availability; the fix is limits, quotas, and capacity.Lockout flood. Login API. Five wrong passwords lock an account, and anyone can send them.Not placed yet.Lockout flood. Login API. Five wrong passwords lock an account, and anyone can send them.Spoofing: Pretending to be someone else. Breaks authentication; the usual fix is stronger proof of identity, such as a second factor or short-lived sessions.Lockout flood. Login API. Five wrong passwords lock an account, and anyone can send them.Repudiation: Doing something and denying it later, with no record to prove otherwise. Breaks accountability; the fix is a log the actor cannot edit.Lockout flood. Login API. Five wrong passwords lock an account, and anyone can send them.Elevation of privilege: Getting rights the user was never given. Breaks authorization; the fix is checking rights on every request, on the server.Lockout flood. Login API. Five wrong passwords lock an account, and anyone can send them.Tampering: Changing data or code without permission. Breaks integrity; the fix is signing or checking what can be changed.Lockout flood. Login API. Five wrong passwords lock an account, and anyone can send them.Information disclosure: Exposing data to someone not meant to see it. Breaks confidentiality; the fix is encryption and saying less.Lockout flood. Login API. Five wrong passwords lock an account, and anyone can send them.Denial of service: Making the service unusable for the people entitled to it. Breaks availability; the fix is limits, quotas, and capacity.Medium. An attacker can lock out every user at once. Fix: slow repeat attempts instead of locking, and limit by source.Support resets an admin. Admin console. The support role can reset any password, including administrators'.Not placed yet.Support resets an admin. Admin console. The support role can reset any password, including administrators'.Spoofing: Pretending to be someone else. Breaks authentication; the usual fix is stronger proof of identity, such as a second factor or short-lived sessions.Support resets an admin. Admin console. The support role can reset any password, including administrators'.Repudiation: Doing something and denying it later, with no record to prove otherwise. Breaks accountability; the fix is a log the actor cannot edit.Support resets an admin. Admin console. The support role can reset any password, including administrators'.Elevation of privilege: Getting rights the user was never given. Breaks authorization; the fix is checking rights on every request, on the server.High. A support login becomes an admin login. Fix: only admins can reset admin passwords.Support resets an admin. Admin console. The support role can reset any password, including administrators'.Tampering: Changing data or code without permission. Breaks integrity; the fix is signing or checking what can be changed.Support resets an admin. Admin console. The support role can reset any password, including administrators'.Information disclosure: Exposing data to someone not meant to see it. Breaks confidentiality; the fix is encryption and saying less.Support resets an admin. Admin console. The support role can reset any password, including administrators'.Denial of service: Making the service unusable for the people entitled to it. Breaks availability; the fix is limits, quotas, and capacity.No limit on reset emails. Login API. Password reset can be requested without limit for any address.Not placed yet.No limit on reset emails. Login API. Password reset can be requested without limit for any address.Spoofing: Pretending to be someone else. Breaks authentication; the usual fix is stronger proof of identity, such as a second factor or short-lived sessions.No limit on reset emails. Login API. Password reset can be requested without limit for any address.Repudiation: Doing something and denying it later, with no record to prove otherwise. Breaks accountability; the fix is a log the actor cannot edit.No limit on reset emails. Login API. Password reset can be requested without limit for any address.Elevation of privilege: Getting rights the user was never given. Breaks authorization; the fix is checking rights on every request, on the server.No limit on reset emails. Login API. Password reset can be requested without limit for any address.Tampering: Changing data or code without permission. Breaks integrity; the fix is signing or checking what can be changed.No limit on reset emails. Login API. Password reset can be requested without limit for any address.Information disclosure: Exposing data to someone not meant to see it. Breaks confidentiality; the fix is encryption and saying less.No limit on reset emails. Login API. Password reset can be requested without limit for any address.Denial of service: Making the service unusable for the people entitled to it. Breaks availability; the fix is limits, quotas, and capacity.Low to medium. It floods users' inboxes and the mail quota. Fix: a limit per address and per hour.

Your sort matched 0 of 8.Your sort matched 1 of 8.Your sort matched 2 of 8.Your sort matched 3 of 8.Your sort matched 4 of 8.Your sort matched 5 of 8.Your sort matched 6 of 8.Your sort matched 7 of 8.Your sort matched 8 of 8.

Spoofing
Pretending to be someone else. Breaks authentication; the usual fix is stronger proof of identity, such as a second factor or short-lived sessions.
Repudiation
Doing something and denying it later, with no record to prove otherwise. Breaks accountability; the fix is a log the actor cannot edit.
Elevation of privilege
Getting rights the user was never given. Breaks authorization; the fix is checking rights on every request, on the server.
Tampering
Changing data or code without permission. Breaks integrity; the fix is signing or checking what can be changed.
Information disclosure
Exposing data to someone not meant to see it. Breaks confidentiality; the fix is encryption and saying less.
Denial of service
Making the service unusable for the people entitled to it. Breaks availability; the fix is limits, quotas, and capacity.

The sort

  • Stolen session cookie: Spoofing. High. The attacker passes as the user. Fix: bind sessions to the device and expire them after an hour idle.
  • Edited login token: Tampering. High. Anything in the token can be rewritten. Fix: sign tokens and reject any that fail the check.
  • Admin denies a deletion: Repudiation. Medium. A deletion cannot be traced to anyone. Fix: an append-only audit log of admin actions.
  • Error reveals the email: Information disclosure. Medium. Attackers learn which emails have accounts. Fix: one message for both cases.
  • Unencrypted backups: Information disclosure. High. One leaked file exposes every password hash. Fix: encrypt backups and restrict who can read them.
  • Lockout flood: Denial of service. Medium. An attacker can lock out every user at once. Fix: slow repeat attempts instead of locking, and limit by source.
  • Support resets an admin: Elevation of privilege. High. A support login becomes an admin login. Fix: only admins can reset admin passwords.
  • No limit on reset emails: Denial of service. Low to medium. It floods users' inboxes and the mail quota. Fix: a limit per address and per hour.

An example

A payments startup was rebuilding its login service before a bank partner's security audit. The first review produced a list of four threats, all variations on stolen passwords, because that was what the team had seen before.

The security lead redrew the service as a data flow: browser, login API, admin console, and database, with a trust boundary wherever data crossed from one owner to another. Then the team walked each part through the six STRIDE types in turn. Spoofing found the stolen session cookie. Tampering found that the login token was encoded but not signed. Repudiation found that admins could delete accounts with no record of who did it. Information disclosure found the error message that told attackers which emails had accounts, and backups sitting unencrypted. Denial of service found that anyone could lock out every user with five wrong passwords each. Elevation found that the support role could reset an administrator's password. Eight threats, ranked by severity, each with a fix and an owner. The audit found nothing the list had missed.

Why it works

People find the threats they have seen before. Asked what could go wrong, a team lists a few vivid attacks and misses whole kinds of failure, because nothing prompts them to look. STRIDE supplies the prompt: six types of threat, each the opposite of a property a secure system needs (authentication, integrity, accountability, confidentiality, availability, and authorization).

Walking every part of the system through every type turns imagination into coverage. The question for each pair is narrow and answerable: can someone pretend to be another user at this component, can someone change this data, can someone act here and deny it. A type with nothing found is either genuinely safe or not yet checked, and the walk makes the difference visible. The data-flow drawing keeps attention on the trust boundaries, where most threats live.

The thinking behind it

Loren Kohnfelder Loren Kohnfelder, a security engineer, devised STRIDE with Praerit Garg at Microsoft in an internal paper of April 1999, The Threats to Our Products. The name is a mnemonic, one letter per kind of threat, from spoofing an identity and tampering with data to denying service and elevating privilege. A short word that holds the whole list makes it hard to skip a kind of threat in a review. Kohnfelder later worked on security and privacy at Google.

Michael Howard Michael Howard, a software security specialist on Microsoft's Security Development Lifecycle team, wrote Writing Secure Code with David LeBlanc (first edition 2001). The book was required reading in Microsoft's 2002 Windows security push, and its second edition (2002) sets out the STRIDE categories. The scheme itself was Kohnfelder and Garg's; Howard and LeBlanc's book is what carried it to a wide public.

Adam Shostack Adam Shostack, a security practitioner who worked on threat modeling at Microsoft, wrote Threat Modeling: Designing for Security (2014), the standard treatment of the practice. As its title says, the book treats threat modeling as a part of designing a system. It sets STRIDE out in full, as one part of a complete method for modeling the threats to a system, and credits STRIDE itself to Kohnfelder and Garg.

How to run it

  1. Draw the system as a data flow. Processes, data stores, outside parties, and the flows between them, on one page.
  2. Mark the trust boundaries. Wherever data passes between owners or privilege levels, such as browser to server or app to database.
  3. Walk each element through the six types. Ask the STRIDE question for each one, and where a type does not apply to an element, say why.
  4. Write each threat as a specific sentence. Who does what to which part, such as a copied cookie logs in as the user.
  5. Rate each threat's severity. High, medium, or low by how much damage and how easy it is; one simple scale applied consistently.
  6. Assign a fix and an owner. Each threat gets a mitigation, an accepted risk with a name on it, or a transfer to someone else.
  7. Rank the list and track it to closure. Review it when the design changes, not only before an audit.

Where it goes wrong

The walk happens without a drawing. The team runs through the six words in the abstract and finds the same threats as before. Draw the flows and boundaries first; the threats are found at specific elements.

The list never closes. Threats are recorded, rated, and left in a document nobody reopens. Put each fix in the same tracker as feature work, with a date.

STRIDE gets treated as a complete security review. It finds design threats, not coding bugs, misconfigured servers, or a supplier's weak spot. Pair it with code review and testing, and with a look at the parts the drawing left out.

When to use something else

  • Red Teaming. When the design is built and the question is whether a skilled adversary can actually break it.
  • Failure Mode and Effects Analysis. When the failures to find are accidents and wear, not attacks, and each needs a severity and likelihood score.
  • Premortem. When the risk is to a project or plan as a whole, imagined as already failed.

Where it sits in SOLVE eX

Go to the source

  • The Threats to Our Products, Microsoft internal memo, 1999
  • Writing Secure Code, second edition, 2002
  • Threat Modeling: Designing for Security, 2014

Back to the catalog