As the network perimeter morphs from physical to virtual, the old Tootsie Pop security model – hard shell on the outside with a soft and chewy center – no longer works. The new mantra, as Mittal Desai, chief information security officer (CISO) at the Federal Energy Regulatory Commission, said at the ATARC CISO Summit: “Never trust, double verify.”
The zero-trust model modernizes conventional network-based security for a hybrid cloud environment. As agencies move systems and storage into the cloud, networks are virtualized and security naturally shifts to users and data. That’s easy enough to do in small organizations, but rapidly grows harder with the scale and complexity of an enterprise.
The notion of zero-trust security first surfaced five years ago in a Forrester Research report prepared for the National Institute for Standards and Technology (NIST). “The zero-trust model is simple,” Forrester posited then. “Cybersecurity professionals must stop trusting packets as if they were people. Instead, they must eliminate the idea of a trusted network (usually the internal network) and an untrusted network (external networks). In zero-trust, all network traffic is untrusted.”
Cloud adoption by its nature is forcing the issue, said Department of Homeland Security Chief Technology Officer Mike Hermus, speaking at a recent Tech + Tequila event: “It extends the data center,” he explained. “The traditional perimeter security model is not working well for us anymore. We have to work toward a model where we don’t trust something just because it’s within our boundary. We have to have strong authentication, strong access control – and strong encryption of data across the entire application life cycle.”
Indeed, as other network security features mature, identity – and the access that goes with it – is now the most common cybersecurity attack vector. Hackers favor phishing and spear-phishing attacks because they’re inexpensive and effective – and the passwords they yield are like the digital keys to an enterprise.
About 65 percent of breaches cited in Verizon’s 2017 Data Breach Investigations Report made use of stolen credentials.
Interestingly however, identity and access management represent only a small fraction of cybersecurity investment – less than 5 percent – according to Gartner’s market analysts. Network security equipment by contrast, constitutes more than 12 percent. Enterprises continue to invest in the Tootsie Pop model even as its weaknesses become more evident.
“The future state of commercial cloud computing makes identity and role-based access paramount,” said Rob Carey, vice president for cybersecurity and cloud solutions within the Global Solutions division at General Dynamics Information Technology (GDIT). Carey recommends creating both a framework for better understanding the value of identity management tools, and metrics to measure that impact. “Knowing who is on the network with a high degree of certainty has tremendous value.”
Tom Kemp, chief executive officer at Centrify, which provides cloud-based identity services, has a vested interest in changing that mix. Centrify, based in Sunnyvale, Calif., combines identity data with location and other information to help ensure only authorized, verified users access sensitive information.
“At the heart of zero-trust is the realization that an internal user should be treated just like an external user, because your internal network is just as polluted as your outside network,” Kemp said at the Feb. 7 Institute for Critical Infrastructure (ICIT) Winter Summit. “You need to move to constant verification.” Reprising former President Ronald Reagan’s “trust but verify” mantra, he adds: “Now it’s no trust and always verify. That’s the heart of zero-trust.”
The Google Experience
When Google found itself hacked in 2009, the company launched an internal project to find a better way to keep hackers out of its systems. Instead of beefing up firewalls and tightening virtual private network settings, Google’s BeyondCorp architecture dispensed with the Tootsie Pop model in which users logged in and then gained access to all manner of systems and services.
In its place, Google chose to implement a zero-trust model that challenges every user and every device on every data call – regardless of how that user accessed the internet in the first place.
While that flies in the face of conventional wisdom, Google reasoned that by tightly controlling the device and user permissions to access data, it had found a safer path.
Here’s an example of how that works when an engineer with a corporate-issued laptop wants to access an application from a public Wi-Fi connection:
- The laptop provides its device certificate to an access proxy.
- The access proxy confirms the device, then redirects to a single-sign-on (SSO) system to verify the user.
- The engineer provides primary and second-factor authentication credentials and, once authenticated by the SSO system, is issued a token.
- Now, with the device certificate to identify the device and the SSO token to identify the user, an Access Control Engine can perform a specific authorization check for every data access. The user must be confirmed to be in the engineering group; to possess a sufficient trust level; and to be using a managed device in good standing with a sufficient trust level.
- If all checks pass, the request is passed to an appropriate back-end system and the data access is allowed. If any of the checks fail however, the request is denied. This is repeated every time the engineer tries to access a data item.
“That’s easy enough when those attributes are simple and clear cut, as with the notional Google engineer,” said GDIT’s Carey, who spent three decades managing defense information systems. “But it gets complicated in a hurry if you’re talking about an enterprise on the scale of the Defense Department or Intelligence community.”
Segmenting the Sprawling Enterprise
A takeaway from 9/11 was that intelligence agencies needed to be better and faster at sharing threat data across agency boundaries. Opening databases across agency divisions, however, had consequences: Chelsea Manning, at the time Pfc. Bradley Manning, delivered a treasure trove of stolen files to WikiLeaks and then a few years later, Edward Snowden stole countless intelligence documents, exposing a program designed to collect metadata from domestic phone and email records.
“The more you want to be sure each user is authorized to see and access only the specific data they have a ‘need-to-know,’ the more granular the identity and access management schema need to be,” Carey said. “Implementing role-based access is complicated because you’ve got to develop ways to both tag data and code users based on their authorized need. Absent a management schema, that can quickly become difficult to manage for all but the smallest applications.”
Consider a scenario of a deployed military command working in a multinational coalition with multiple intelligence agencies represented in the command’s intelligence cell. The unit commands air and ground units from all military services, as well as civilians from defense, intelligence and possibly other agencies. Factors determining individual access to data might include the person’s job, rank, nationality, location and security clearance. Some missions might include geographic location, but others can’t rely on that factor because some members of the task force are located thousands of miles away, or operating from covert locations.
That scenario gets even more complicated in a hybrid cloud environment where some systems are located on premise, and others are far away. Managing identity-based access gets harder anyplace where distance or bandwidth limitations cause delays. Other integration challenges include implementing a single-sign-on solution across multiple clouds, or sharing data by means of an API.
Roles and Attributes
To organize access across an enterprise – whether in a small agency or a vast multi-agency system such as the Intelligence Community Information Technology Enterprise (IC ITE) – information managers must make choices. Access controls can be based on individual roles – such as job level, function and organization – or data attributes – such as type, source, classification level and so on.
“Ultimately, these are two sides of the same coin,” Carey said. “The real challenge is the mechanics of developing the necessary schema to a level of granularity that you can manage, and then building the appropriate tools to implement it.”
For example, the Defense Department intends to use role-based access controls for its Joint Information Enterprise (JIE), using the central Defense Manpower Data Center (DMDC) personnel database to connect names with jobs. The available fields in that database are in effect, the limiting factors on just how granular role-based access controls will be under JIE.
Access controls will only be one piece of JIE’s enterprise security architecture. Other features, ranging from encryption to procedural controls that touch everything from the supply chain to system security settings, will also contribute to overall security.
Skeptical of Everything
Trust – or the lack of it – plays out in each of these areas, and requires healthy skepticism at every step. Rod Turk, CISO at the Department of Commerce, said CISOs need to be skeptical of everything. “I’m talking about personnel, I’m talking about relationships with your services providers,” he told the ATARC CISO Summit. “We look at the companies we do business with and we look at devices, and we run them through the supply chain. And I will tell you, we have found things that made my hair curl.”
Commerce’s big push right now is the Decennial Census, which will collect volumes of personal information (PI) and personally identifiable information (PII) on almost every living person in the United States. Conducting a census every decade is like doing a major system reset each time. The next census will be no different, employing mobile devices for census takers and for the first time, allowing individuals to fill out census surveys online. Skepticism is essential because the accuracy of the data depends on the public’s trust in the census.
In a sense, that’s the riddle of the whole zero-trust concept: In order to achieve a highly trusted outcome, CISOs have to start with no trust at all.
Yet trust also cuts in the other direction. Today’s emphasis on modernization and migration to the cloud means agencies face tough choices. “Do we in the federal government trust industry to have our best interests in mind to keep our data in the cloud secure?” Turk asked rhetorically.
In theory, the Federal Risk and Authorization Management Program (FedRAMP) establishes baseline requirements for establishing trust but doubts persist. What satisfies one agency’s requirements may not satisfy another. Compliance with FedRAMP or NIST controls equates to risk management rather than actual security, GDIT’s Carey points out. They’re not the same thing.
Identity and Security
Beau Houser, CISO at the Small Business Administration, is more optimistic by improvements he’s seen as compartmentalized legacy IT systems are replaced with centralized, enterprise solutions in a Microsoft cloud.
“As we move to cloud, as we roll out Windows 10, Office 365 and Azure, we’re getting all this rich visibility of everything that’s happening in the environment,” he said. “We can now see all logins on every web app, whether that’s email or OneDrive or what have you, right on the dashboard. And part of that view is what’s happening over that session: What are they doing with email, where are they moving files.… That’s visibility we didn’t have before.”
Leveraging that visibility effectively extends that notion of zero-trust one step further, or at least shifts it into the realm of a watchful parent rather than one who blindly trusts his teenage children. The watchful parent believes trust is not a right, but an earned privilege.
“Increased visibility means agencies can add behavioral models to their security controls,” Carey said. “Behavioral analysis tools that can match behavior to what people’s roles are supposed to be and trigger warnings if people deviate from expected norms, is the next big hurdle in security.”
As Christopher Wlaschin, CISO at the Department of Health and Human Services, says: “A healthy distrust is a good thing.”