Wireless communication systems are inherently vulnerable to intentional jamming. In this paper, two classes of such jammers are considered: Those with partial and full knowledge. While the first class accounts for those jammers that know the encoding and decoding function, the latter accounts for those that are further aware of the actual transmitted message. Of particular interest are so-called denial-of-service (DoS) attacks in which the jammer is able to completely disrupt any transmission. Accordingly, it is of crucial interest for the legitimate users to detect such adversarial DoS attacks. This paper develops a detection framework based on Turing machines. Turing machines have no limitations on computational complexity and computing capacity and storage and can simulate any given algorithm. For both scenarios of a jammer with partial and full knowledge, it is shown that there exists no Turing machine which can decide whether or not a DoS attack is possible for a given channel and the corresponding decision problem is undecidable. On the other hand, it is shown for both scenarios that it is possible to algorithmically characterize those channels for which a DoS attack is not possible. This means that it is possible to detect those scenarios in which the jammer is not able to disrupt the communication. For all other channels, the Turing machine does not stop and runs forever making this decision problem semidecidable. Finally, it is shown that additional coordination resources such as common randomness make the communication robust against such attacks.