Removing token attribute from Authorizations API responses

Since OAuth access tokens function like passwords, they should be treated with care. Today we are making it easier to more securely work with authorizations via the Authorizations API. We are deprecating the use use of the token attribute in the majority of the Authorizations API responses. For the affected APIs, the token attribute will soon return an empty string. To get ready for that change, we are giving developers a chance to preview the updated API starting today.

What’s changing?

The current OAuth Authorizations API requires GitHub to store the full value for each OAuth token on our servers. In order to increase the security for our users, we are changing our architecture to store the SHA-256 digest of OAuth tokens instead. GitHub securely hashes user passwords using bcrypt and we want to provide comparable security for OAuth tokens as well.

Rest assured that this change is an entirely proactive measure from GitHub and is not associated with any security incident.

Who is affected?

This change affects any code that relies on accessing the token attribute from these OAuth Authorizations API responses. For example, our own GitHub for Mac and GitHub for Windows applications relied on reading the token from the Get-or-create an authorization for a specific app API, in order to support multiple installations of our desktop application for a single user.

What should you do?

In order to reduce the impact of removing the token attribute, the OAuth Authorizations API has added a new request attribute (fingerprint), added three new response attributes (token_last_eight, hashed_token, and fingerprint), and added one new API. While these new APIs and attributes do not replace the full functionality that previously existed, they can be used in place of token for most common use cases.

• token_last_eight returns the last eight characters of the associated OAuth token. As an example, token_last_eight could be used to display a list of partial token values to help a user manage their OAuth tokens.

• hashed_token is the base64 of the SHA-256 digest of the token. hashed_token could be used to programmatically validate that a given token matches an authorization returned by the API.

• fingerprint is a new optional request parameter that allows an OAuth application to create multiple authorizations for a single user. fingerprint should be a string that distinguishes the new authorization from others for the same client ID and user.

  For example, to differentiate installations of a desktop application across multiple devices you might set fingerprint to SHA256_HEXDIGEST("GitHub for Mac - MAC_ADDRESS_OF_MACHINE"). Since fingerprint is not meant to be a user-facing value, you should still set the note attribute to help a user differentiate between authorizations on their OAuth applications listing on GitHub.

• Get-or-create an authorization for a specific app and fingerprint is a new API that is analagous to the Get-or-create an authorization for a specific app API, but adds support for the new fingerprint request parameter.

Preview period

We are making the new Authorizations API available today for developers to preview. During this period, we may change aspects of these endpoints. If we do, we will announce the changes on the developer blog, but we will not provide any advance notice.

While these new APIs are in their preview period, you’ll need to provide the following custom media type in the Accept header:

application/vnd.github.mirage-preview+json

We expect the preview period to last 4-6 weeks. (Stay tuned to the developer blog for updates.) At the end of the preview period, these changes will become an official and stable part of GitHub API.

Migration period

At the end of the preview period, we will announce the start of a migration period. Developers will have 8 weeks to update existing code to use the new APIs.

Why SHA-256 over bcrypt?

Some users may be curious why we are not using bcrypt to hash our OAuth tokens like we do for user passwords. Bcrypt is purposefully computationally expensive in order to mitigate brute force attacks against low entropy passwords. However, OAuth tokens are highly random and are not susceptible to brute force attacks. Given that OAuth token validation occurs for each request to the API we chose SHA-256 for performance reasons.

If you have any questions or feedback, please drop us a line.