A token (in the context of authentication) is a piece of data that contains information about the entity to be authenticated. By presenting a token during authenticated requests, the receiving party can verify claims and information about the entity that the request is performed on behalf of before responding or granting access to a resource.
An advantage of using a token is that the user never has to directly present their private credentials or passwords to the resource server. Instead, the user will have to log in to a trusted third-party that an organization delegates to, i.e. an identity provider. The identity provider will issue a token that the client application can use to authenticate itself to a resource server on behalf of a user, if any.
The token itself can be self-contained, meaning that it contains all the information necessary for a resource server to make assertions about the entity that the token is associated with. Additionally, the token will often have an expiry time to limit its use to a typical user session.
Most references to the word "token" in our case actually refers to a self-contained JSON Web Token (JWT). Though it should be noted that a JWT is a token, but not all tokens are JWTs.
JSON Web Token (JWT) is a compact, self-contained means of representing claims as a JSON object to be transferred between two parties, as specified in RFC 7519.
The contained claims can be verified and trusted by validating its digital signature.
In most cases, the JWT is signed using an asymmetric secret, i.e. public and private key pair:
- The sending party will sign the JWT using their private key.
- The receiving party may verify the signature using the signer's associated public key, which may be distributed out-of-band or published at a well-known endpoint.
A JWT will look something like this in its serialized, compact form (where each part is base64-URL-encoded):
The token consists of three parts:
The header consists of metadata describing the token, such as:
kid: The key ID of the associated key used to sign the token.
typ: The type of token, usually
alg: The algorithm used to sign the token, usually
The payload contains claims about the associated user or principal that the token represents as well as additional data.
JWTs will mostly consist of registered claims or "standard" claims. Depending on the identity provider however, additional claims may be included.
Example (with registered claims only):
See RFC 7519, section 4.1 for descriptions of these claims.
Parties receiving JWTs should validate the claims found in the payload. See token validation.
While the RFC does not require a JWT to contain a signature, a JWT will quickly lose its value if it does not have a signature at all. The signature is used to verify the integrity of the information contained within the token. Any recipient of the JWT can verify that the token, nor its contents, has been tampered with or modified at any point after being issued by the issuer.
Signatures are created by using the header and payload as input. Use a library to do this instead of implementing it yourselves, see best practices.
Interested readers may find a more thorough introduction on JWTs here:
And, of course (for the adventurous), the RFC itself:
An ID token is only included as part of the OpenID Connect standard. It is required to be a JWT. The ID token is strictly used for authentication of end-users. It should never be sent to a resource server (an API server).
The ID token only asserts that a user has been logged in at the identity provider and authenticated, and that the intended recipient is the client application that initiated the login.
However, the ID token itself should not grant access to any resource on behalf of the end-user. This is because the
intended recipient of the token (the
aud or audience claim) is equal to the client application, not the resource
server. The correct token to use in such cases is an access token.
An access token is a token that allows the client application to perform authenticated requests on behalf of the end-user. In OAuth 2.0 terms, this is what is known as delegated authorization.
Access tokens are often in the form of JWTs as they're self-contained, which allows for local validation without the need for network calls.
All the identity providers we use will issue access tokens in the form of JWTs.
More on ID Token vs Access Token
Auth0 has a great blog post on this: https://auth0.com/blog/id-token-access-token-what-is-the-difference/
And an accompanying video if you prefer: https://www.youtube.com/watch?v=M4JIvUIE17c
A refresh token is an opaque string (only meaningful for the recipient, i.e. an identity provider) that may be included as part of the token set returned when performing authentication flows in OpenID Connect.
The token can be used to obtain additional access tokens from the identity provider for the token's associated user session in order to extend a user's session whenever the original access token is no longer valid.
Beware though, a refresh token is quite powerful as it allows an application to obtain access tokens for a user as long as the refresh token is valid. Thus, the refresh token should be stored securely (server-side) and not be used for prolonged periods.
As best practice, you should expire a session if:
- The user has been authenticated for some longer period of time, e.g. a typical workday of 8 hours or so
- The user has been inactive for some time, e.g. 15 - 30 minutes.
The user should ideally be redirected to the identity provider to ensure that the user can still identify themselves, if necessary. This is to remediate potential abuse from any adversaries that may have shared access to the user client, for example when the user has forgotten to log out and is no longer present at the device.
Tokens should be validated before granting access to resources. There are two distinct cases where your client should perform such validation:
- Your client is a resource server / API that receives authenticated requests from consumers (RFC 6749, Section 7)
- Your client performs the OpenID Connect Authorization Code Flow (OpenID Connect Core 1.0, Section 18.104.22.168)
In both cases, we have a set of self-contained JWTs (both ID Tokens and Access Tokens) that should be validated. Validation involves the following steps:
- Parsing and decoding the JWT
- Validating the signature and its algorithm
- Validating the claims in the payload
Validation must be performed before granting access to any resource server / API.
Access Tokens must always be validated by the Resource Server / API before granting access. Clients should normally just pass the access token along to the resource server without any processing of it, however if any processing is performed, clients must also perform such validation.
A JWT usually contains the
kid (key ID) claim in the token's header to indicate which key was used to sign
the token. The signature can then be verified by finding the matching public key from the
For our providers, the algorithm should be at least
RS256. If the algorithm is set to
none or the JWT is not signed,
the request should be rejected.
ID Tokens have a defined set of standard claims and steps required for validation of these in the OIDC Core specification, Section 22.214.171.124.
Access Tokens has an equivalent specification for validation in RFC 9068. However, many of our providers do not fully support this specification as it is relatively new.
The most important claims to validate are the following:
- The principal that is the subject of the token. Usually an identifier for the end-user or client that requests a resource.
- The timestamp for which the token expires, in unix time (seconds since epoch). The token is invalid if it is used afer this time.
nbf(not before time)
- The timestamp for which the token is considered valid, in unix time (seconds since epoch). The token is invalid if it is used before this time.
iat(issued at time)
- The timestamp for which the token was issued, in unix time (seconds since epoch). This can be used to determine the age of the token.
If any of these claims are missing or contain unexpected values, validation should fail and the token should be rejected.
In most cases, you should not parse and validate tokens manually. It is strongly recommended that you use well-known and widely used libraries and frameworks that takes care of most of the heavy lifting (parsing, validation of signature and standard claims) for you.
See libraries and frameworks for a non-comprehensive list.