On the security of modern Single Sign-On Protocols: Second-Order Vulnerabilities in OpenID Connect

OAuth is the new de facto standard for delegating authorization in the web. An important limitation of OAuth is the fact that it was designed for authorization and not for authentication. The usage of OAuth for authentication thus leads to serious vulnerabilities as shown by Zhou et. al. in [44] and Chen et. al. in [9]. OpenID Connect was created on top of OAuth to fill this gap by providing federated identity management and user authentication. OpenID Connect was standardized in February 2014, but leading companies like Google, Microsoft, AOL and PayPal are already using it in their web applications [1], [2], [3], [30]. In this paper we describe the OpenID Connect protocol and provide the first in-depth analysis of one of the key features of OpenID Connect: the Discovery and the Dynamic Registration extensions.We present a new class of attacks on OpenID Connect that belong to the category of second-order vulnerabilities. These attacks consist of two phases: First, the injection payload is stored by the legitimate application. Later on, this payload is used in a security-critical operation. Our new class of attacks - called Malicious Endpoints attacks - exploits the OpenID Connect extensions Discovery and Dynamic Registration. These attacks break user authentication, compromise user privacy, and enable Server Side Request Forgery (SSRF), client-side code injection, and Denial-of-Service (DoS). As a result, the security of the OpenID Connect protocol cannot be guaranteed when these extensions are enabled in their present form. We contacted the authors of the OpenID Connect and OAuth specifications. They acknowledged our Malicious Endpoint attacks and recognized the need to improve the specification [29]. We are currently involved in the discussion regarding the mitigation of the existing issues and an extension to the OAuth specification.