Asymmetric fingerprinting protocols are designed to prevent an untrustworthy Provider incriminating an innocent Buyer. These protocols enable the Buyer to generate their own fingerprint by themself, and ensure that the Provider never has access to the Buyer's copy of the Work. Until recently, such protocols were not practical because the collusion-resistant codes they rely on were too long. However, the advent of Tardos codes means that the probabilistic collusion-resistant codes are now sufficiently short that asymmetric fingerprint codes should, in theory, be practical. Unfortunately, previous asymmetric fingerprinting protocols cannot be directly applied to Tardos codes, because generation of the Tardos codes depends on a secret vector that is only known to the Provider. This knowledge allows an untrustworthy Provider to attack traditional asymmetric fingerprinting protocols. We describe this attack, and then propose a new asymmetric fingerprinting protocol, specifically designed for Tardos codes.