What You Should Know About Tokenization

A week ago Visa released a set of best practices and recommendations for tokenization. Unfortunately, "best practices" leaves plenty of room for poor implementations.A few months back I wrote a post about token deployment strategies for meeting PCI compliance. What I did not discuss were some of the differences between the different tokenization technologies on the market.

Token solutions have become popular because they remove credit card data from most processing systems, thus eliminating them from inspection during PCI assessment. For example, if you have a dozen systems (order entry, customer management, payment gateways, general ledger, etc.) and you substitute a token for the Primacy Account Number, then you remove a huge portion of the PCI audit. For a lot of merchants, that means a savings of 50 percent. No credit card numbers, no security threat, so no reason to poke around.

But that assumes the token is secure. The critical part of a token strategy is to ensure the token does not betray the original credit card number. Tokens created via any mathematical function, be it cryptography or hashing, always start with the account number. That means there is a chance they can be reversed back into the original if not carefully implemented or deployed. But we know from experience that poorly implemented algorithms, bad entropy or pseudo-random number generators, or improper use of padding/salting results in tokens that are easy to hack. The only two recommendations made by Visa are for mathematical derivatives, and there is considerable leeway in its guidance. In other words, a solution that meets Visa's criteria can provide poor security.

What does this mean to you? Several things:

1. Visa should have included in its recommendation the use of completely random numbers. This is far more secure because there is simply no way to reverse-engineer the credit card number from the token given there is no mathematical relationship. The only way to gain access to the original data is through the token server itself. I recommend you select this option if it is available from your vendor.

2. If you are looking at a solution that uses cryptographic functions, then you need to understand you will be using some form of a format-preserving encryption to form the token. Despite being based on accepted strong cryptographic algorithms, the format-preserving options are not specifically endorsed by Visa or the PCI Standards Council. Make sure your vendor has had its product professionally reviewed by a noted expert in the field of cryptanalysis. Also, verify that your auditor will remove systems using encryption from the scope of the audit -- otherwise you miss out on cost savings.

3. If you are looking at a solution that uses a hashing variant, then first make sure the method used is acceptable to Visa and PCI. Second, verify that the vendor implementation has been reviewed by the cryptanalysis community. Finally, see if you can locate a product that provides random salt values for each token. Static salt values or salting with a finite set of merchant IDs offers poor security and makes the hashes vulnerable to dictionary attacks.

Take the time to verify these options so you can get full value for your tokenization investment.

Adrian Lane is an analyst/CTO with Securosis LLC, an independent security consulting practice. Special to Dark Reading.