Abstract. Java Cards have been threatened so far by attacks using ill-formed applications which assume that the application bytecode is not verified. This assumption remained realistic as long as the bytecode verifier was commonly executed off-card and could thus be bypassed. Nevertheless it can no longer be applied to the Java Card 3 Connected Edition context where the bytecode verification is necessarily performed on-card. Therefore Java Card 3 Connected Edition seems to be immune against this kind of attacks. In this paper, we demonstrate that running ill-formed application does not necessarily mean loading and installing ill-formed application. For that purpose, we introduce a brand new kind of attack which combines fault injection and logical tampering. By these means, we describe two case studies taking place in the new Java Card 3 context. The first one shows how ill-formed applications can still be introduced and executed despite the on-card bytecode verifier. The second example leads to the modification of any method already installed on the card into any malicious bytecode. Finally we successfully mount these attacks on a recent device, emphasizing the necessity of taking into account these new threats when implementing Java Card 3 features.

Until 2009, Java Cards have been mainly threatened by Logical Attacks based on ill-formed applications. The publication of the Java Card 3.0 Connected Edition specifications and their mandatory on-card byte code verification may have then lead to the end of software-based attacks against such platforms. However, the introduction in the Java Card field of Fault Attacks, well-known from the cryptologist community, has proven this conclusion wrong. Actually, the idea of combining Fault Attacks and Logical Attacks to tamper with Java Cards appears as an even more dangerous threat. Although the operand stack is a fundamental element of all Java Card Virtual Machines, the potential consequences of a physical perturbation of this element has never been studied so far. In this article, we explore this path by presenting both Fault Attacks and Combined Attacks taking advantage of an alteration of the operand stack. In addition, we provide experimental results proving the practical feasibility of these attacks and illustrating their efficiency. Finally, we describe different approaches to protect the operand stack's integrity and compare their cost with a particular interest on the time factor.