Supply Chain: Finding Bad DevicesResearch Looks at Suspicious Behavior in Network Devices
Researchers at the Georgia Institute of Technology are working on a way to profile devices along the supply chain to identify whether they've been compromised.
Auditing for compromised devices along the supply chain can be a very expensive process, says Paul Royal, associate director of the Georgia Tech Information Security Center. "If you look on the network, a device will announce itself as a thermostat but may actually be a laptop, but that laptop may not have all the properties," he says in an interview with Information Security Media Group (see transcript below). Royal says discovering those compromised devices can be time-consuming.
That's why Georgia Tech researchers are working on a way to profile a device. The process starts with someone booting up a given system and observing the traffic it generates, according to Georgia Tech's recently issued Emerging Cyber Threats Report 2014. Using network probes, the individual can determine if the systems' architecture is compromised and a device on the system is not what it purports to be.
Royal, in the interview, also addresses:
- An application called CloudCapsule that encrypts data on mobile and other devices before information is securely transmitted and stored in the cloud;
- Continuing work on a method for enabling a search of encrypted data stored in the cloud; and
- Steps researchers take to ensure that any newly developed security solutions can be easily applied.
Royal is a research scientist in the College of Computing at Georgia Tech, where he engages in collaborative research on various facets of the online criminal ecosystem. Earlier, Royal served as principal researcher at Purewire, where he identified emerging threats and designed methods that enhanced the company's web security service.
ERIC CHABROW: The Georgia Institute of Technology just issued its Emerging Cyber Threats Report 2014, and it addresses some of the critical IT security challenges organizations face. There's a lot of fascinating cybersecurity research going on at Georgia Tech that addresses those challenges. Let's talk about several of those research projects. One that caught my attention is known as CloudCapsule. What is CloudCapsule and why is it needed?
PAUL ROYAL: I'll actually start by providing some motivating background to the "why." What we've seen as a development within the mobile space as it intersects with enterprises is obviously BYOD - bring-your-own-device - and that's effectively dissolving the network perimeter. It really means to use your device that you bring for various work-related purposes. Inevitably, work data is going to be placed onto that device and then, depending on what cloud services you integrate on your device, data that has been placed on the device will in turn end up in the cloud. Obviously, we want to protect this data.
One idea we had is characterized as CloudCapsule, which is basically a way of entering what we characterize as a secure mode on a mobile device and then being able to place data into a cloud service that's transparently encrypted. ...
The idea is that within mobile operating systems - and mobile operating systems are a little bit different than desktop operating systems because they were designed in an era after people had begun to think about security, which means that we consider security in the design of an operating system - instead of putting it on after the fact, we can include it from the ground up. Mobile operating systems ... or at least a subset of them, have good process isolation. We want to take that isolation a step further, which means providing some level of transparent protection. In this case, that protection is encryption, going into a secure mode and saying, "I'd like to make these files available from various end-user devices, etc." When you place into that folder, they're seamlessly or transparently encrypted. When you retrieve them from the cloud, they're decrypted.
Use of Encryption
CHABROW: The encryption is occurring on the cloud, the device, or both?
ROYAL: It's occurring on the device prior to delivery of that data into the cloud. By the time the data hits the cloud service, it's encrypted. I think it's important to differentiate where the encryption occurs because if the encryption occurs at the cloud-service-provider level, there are obviously two risks. One of them is criminals who would break into the cloud service to compromise or to obtain the data, and the second issue is a concern about a sovereign government relative to its authority over business to basically request the keys used by a cloud provider to encrypt data. For those reasons, we feel that it's best that the encryption occur before the data leaves the device and goes into the cloud at the device level. They would download an app - the app would represent CloudCapsule - and then through that app they would be able to put data into the cloud and retrieve it from the cloud.
CHABROW: Is this working already?
ROYAL: We have a version for iOS and you can request a copy of it.
CHABROW: Are you going to make it available for other platforms?
ROYAL: Yes, that's our intention. Different platforms have different features. With respect to our ability to use those features to secure a device, additional work would obviously be needed. In some cases, for example, if we were to release a desktop version of CloudCapsule, we could probably make stronger security guarantees because one advantage we have on the desktop is it's relatively mature and ubiquitous virtualization. In this case, instead of process isolation, we can use the hypervisor to increase isolation to protect that data.
Searching in the Cloud
CHABROW: Let's talk about searching for information in the cloud. This is where the report points out encryption can secure data stored on the online service, but focusing on security alone can rob the company of much of the ability to use the data. I want to discuss the research you're conducting to narrow that tradeoff between security functionality and efficiency regarding search.
ROYAL: That's an important segway into what we were just discussing. If you encrypt the data traditionally, by decrypting you were unable to search over the resulting encrypted data. For some people, that's not acceptable. In short, there's a desire to be able to, to varying degrees, index data to perform basic keyword searches over that data and to sort that data, even if it's encrypted.
There are some cryptographers at Georgia Tech and specifically those affiliated with the Georgia Tech Information Security Center that have come up with algorithms whereby the data, even in encrypted form, has certain properties that permit basic keyword search. This isn't, for example, the addition of plain-text metadata. This is actually being able to search over encrypted data. Effectively, what that means is that, when you search, your criteria is likewise encrypted and the encrypted representation of your basic keyword search is used over that data. The encryption algorithm was designed in a way so that search is possible.
CHABROW: Would this be all encrypted data in the database, or certain types of data?
ROYAL: Certain types and within certain limitations. Obviously, there are going to be tradeoffs between the functionalities offered, the security offered and the efficiency. That probably forms something of a triangle where if you get more of one, you have less of the other two, or perhaps more of two and less of one.
CHABROW: Is there a term used for this encrypted search?
ROYAL: Sometimes we refer to it as searchable encryption. There are stronger ways of representing it. At least there are other types, like order-preserving encryption.
CHABROW: What is that?
ROYAL: The ability to order the data has been preserved after it has been encrypted. Unfortunately, that definitely comes at a cost of some security because the more the encrypted data is capable of revealing, as the result of the functionalities or the utility that it possesses in encrypted form, obviously the more it may reveal to an adversary who has stolen that data.
How Encrypted Search Works
CHABROW: The cryptography that you're using or the algorithms that you're developing for this searchable encryption, would it have to correspond with the encrypted data itself? In other words, when you encrypt the data that's being searched, there are some tags. When an encrypted search engine is used, it could be available to be screened?
ROYAL: ... Again, the traditional way of searching encrypted data is to apply unencrypted metadata, and that's not what we're doing here. We're integrating the preservation of certain properties of the original data within the encrypted data so that you can take advantage of those properties; for example, to do a keyword search without revealing the contents of the encrypted data to someone who may have acquired that data without authorization.
CHABROW: Once something is identified, how do people see it?
ROYAL: Let's say they do a basic keyword search. The keyword would be encrypted. The encrypted keyword would be searched over the data. The result set would be a set of encrypted data. That encrypted data would be served to an end device. Once it hits the end device, the encrypted data is decrypted and the user sees the plain text representation of the results.
CHABROW: What's the status of this?
ROYAL: The designs of any of these algorithms are actually complete and a subset of it has already been published in academic conferences.
CHABROW: Do you see this as a commercial product?
ROYAL: There have been tech-transfer efforts at Georgia Tech in the past and there's some interest from industry, but obviously by publishing on it we hope that those who need it will adopt it.
Identifying Manipulated Devices
CHABROW: One of the problems with securing the supply chain is determining whether each piece of equipment linked to the network isn't manipulated along the chain to open systems to vulnerabilities. That would be just too time-consuming to go over each piece and to see whether there are vulnerabilities there or not. Tell us about the research your scientists are pursuing. You call it fingerprinting and profiling devices on the network using the traffic that moves to and from the devices as a way to vet these devices' security.
ROYAL: As it pertains to interactions in the network, devices will behave differently. In many cases, it can be difficult to successfully forward or fabricate all behaviors of a device that you're pretending to be. Concisely, if you look on the network, a device will announce itself as a thermostat but may actually be a laptop, but that laptop may not have all the properties. The responses would be expected timings and so forth of a thermostat, and those can be telltale signs that, in fact, something is wrong and that there's an issue, for example, with a network relative to the devices that claim to be on the network.
CHABROW: Is it limited to just one type of device mimicking another?
ROYAL: There's potential for things like hardware root kits. It's very costly to actually examine every integrated circuit or what have you on a given piece of electronic [equipment]. Heuristics, for example, as simple as power draw can be a telltale sign that maybe this device is a genuine article, but something has been added to it and what has been added has, unfortunately for the adversary but good for the defender, a measurable side effect.
CHABROW: What's the status of this research?
ROYAL: This research is ongoing because supply chain integrity continues to be a concern. There are many open problems still within the space; for example, if you use a supplier and something that's compromised physically may come into your door. Be diligent in your monitoring. If, for example, you have a networking switch that has in some way been compromised, monitor the network behavior that switched to see if at some point it's going to exhibit behavior that's not expected or is an indicator of compromise.