In most cases, trust boundaries are violated when a secret is exposed in a source code repository or an uncontrolled deployment environment.
Unintended people who don’t need to know the secret might get access to it. They might then be able to use it to gain unwanted access to associated
services or resources.
The trust issue can be more or less severe depending on the people’s role and entitlement.
What is the potential impact?
Azure provide the ability to deploy models to online HTTP REST endpoints for use in real-time inferencing. Real-time inferencing is the process of
using a deployed model and other transformations logics to predict values in real-time.
To access these endpoints, a user must authenticate with the service, using API keys, AML Tokens or Mantra ID.
Hardcoding an API key in code allows an attacker to freely use the service without any restrictions, while business logic might assume that the
access is restricted to the owner of the API key.
Below are some real-world scenarios that illustrate some impacts of an attacker exploiting the key.
Exceeding rate limits
Using a leaked secret, an attacker may be able to make hundreds or thousands of authenticated calls to an online service. It is common for online
services to enforce a rate limit to prevent their servers from being overwhelmed.
If an attacker is able to exceed a user-based rate limit, they may be able to cause a denial of service for the user. If this continues over a long
period of time, the user may also be subject to additional fees or may have their account terminated.
Financial loss
Financial losses can occur when a secret is used to access a paid third-party-provided service and is disclosed as part of the source code of
client applications. Having the secret, each user of the application will be able to use it without limit to use the third party service to their own
need, including in a way that was not expected.
This additional use of the secret will lead to added costs with the service provider.
Moreover, when rate or volume limiting is set up on the provider side, this additional use can prevent the regular operation of the affected
application. This might result in a partial denial of service for all the application’s users.
