As I mentioned in my last blog, one of the sessions I gave recently at RSA Conference China was a discussion of “Keys and Clouds”, exploring various models for key management and encryption in the cloud. It’s a topic that comes up often in my meetings with customers about private, public and hybrid cloud strategy. It’s also something that we’ve been giving a lot of thought to in the OASIS KMIP (Key Management Interoperability Protocol) standards committee that I co-chair. In fact, we’ll be exploring the cloud-related use cases in our KMIP face-to-face next week and also discussing them in the KMIP webinar we’ll be giving later in September.
In the KMIP committee, we’ve focused on three fundamental models for managing keys for encrypted information entrusted to or processed in the cloud. The first model retains both encryption and key management in the enterprise, entrusting only encrypted information to a Cloud Service Provider (CSP). Many cloud data storage, backup and archive offerings support this approach.
The second model retains key management in the enterprise but distributes keys to the CSP as needed.
The CloudLink® offering from Afore is a good example of this.
The third model entrusts key management as well as encryption to the cloud service provider, a model that is not very common at the moment, though the public information on Microsoft® Azure™ Trust Services indicates interest in this approach.
Defining these three models has been important in our KMIP work so that we can look at the implications for the protocol and the environments in which it’s used. For example, both the second and third model imply mechanisms by which the CSP can distinguish among tenants, impacting client registration processes and client-related objects and attributes in the protocol. The models also imply tenant-specific administrative interfaces that may need to be supported in the protocol rather than provided as proprietary capabilities by the key management vendors. These are the kinds of issues we’ll be exploring in our face-to-face meeting.
These issues are a subset of larger questions about cross-domain interactions and other new challenges for cryptographic capabilities. NIST has been exploring these larger issues in a couple of new draft publications and has organized a workshop to review their work next week in Washington. Colleagues from KMIP and I will be at the workshop and will be leading one of the panels, talking not only about cloud but more generally about key management issues when supporting interactions across security domains, including with both well-known technologies like Hardware Security Modules and relatively new ones like Quantum Key Distribution.
These explorations of key management technology are very important in terms of establishing effective support for our complex IT environments and effective responses to the sophisticated attacks we face now. But in themselves they don’t directly answer the question of how an enterprise should do key management in the light of these complexities and attacks. It’s helpful to use cost/benefit analysis in considering the alternatives. But cost/benefit doesn’t say much about the various approaches regarding their relative vulnerability and attractiveness to attackers. Does consolidation of key management at a CSP increase the risk of compromise? Or does it increase the security of keys through more comprehensive control and visibility technologies, processes and expertise that a CSP may be able to put in place, compared to individual enterprises?
One of the tools that is helpful in exploring this question is game theory. As I discussed in “The Game of Cybersecurity”, my colleagues in RSA labs have been exploring the intersection of security and game theory for some time, including in their seminal work with Ron Rivest on modeling advanced attacks through the FlipIt game. We’ve just had a paper accepted for the Gamesec conference in November that will explore the application of FlipIt in a number of security decisions, such as password reset strategies and key management deployment models. I’ll also be giving a session on “Games of Keys and Clouds” at RSA Conference Europe that will discuss the intersection of game theory and key management. You can find a podcast in which Jeanne Friedman and I talk about that session on the RSA Conference Europe site.
There’s still lots of work to be done in developing tools and providing information to help enterprises make effective decisions regarding keys and clouds. But the webinars and conferences I’ve mentioned above will at least provide some useful insights and guidance. I hope you can join us at one or more of them!