Why Security Fails

In thinking about how nuclear security and safeguards can fail, it is useful to keep in mind why security usually fails in general.  Most security managers and organizations have a good understanding of the assets they are trying to protect, the resources available to them to protect those assets, and the consequences should security fail (though this is sometimes greatly underestimated).  They often have a reasonably accurate understanding of the threats they face—who might attack, why, how, when, and with what goals and resources.  What is often lacking is a good understanding of the vulnerabilities—the weaknesses in the security that can be exploited by the threats—and how those vulnerabilities can be mitigated or eliminated.

Some recent major security failures serve as good examples.  The intrusion deep into the Y-12 nuclear facility by an 82-year old nun and her fellow protesters wasn’t because nuclear protesters weren’t understood as a threat for trespassing.  Similarly, the recent incident at the White House where an individual jumped the fence and entered the unlocked front door of the White House is another example where the threat was well understood—people have been jumping the fence at the White House for a long time—but the security vulnerabilities were either unrecognized or not properly neutralized.  The damaging cyber attacks at Target and Sony were similar in that the threat was not new, but the vulnerabilities were poorly recognized and dealt with.

Much of the problem in my view is that the vulnerability assessment (VA) component of risk management—especially for nuclear security/safeguards and especially for layered security—is largely missing in action.  Nuclear facilities and organizations will frequently claim to analyze their security, but these assessments are often highly inadequate for understanding vulnerabilities, including those that can be exploited by insiders.

A number of different security analysis techniques—while still potentially useful—frequently get confused with VAs.  These include security surveys (walking around with a checklist), security audits (checking if the rules are being followed), comparing security practices with general “standards” or “guidelines” or “best practices” (which may themselves be flawed or too simplistic or too generalized for the local culture).  Other techniques not very effective at finding vulnerabilities include threat assessments, feature analyses, fault or event tree analyses (from safety engineering; often very problematic for security analysis), the Delphi Method (a technique for getting a decision from a panel of experts), software assessment tools, the CARVER Method (often used by the U.S. Department of Defense and law enforcement), and Modern Risk Management. 

The widely used Design Basis Threat (DBT)—the common sense idea that security should be designed to counter real threats—is particularly bad at identifying vulnerabilities because it is a threat analysis technique.  The common practice of using DBT to “test” nuclear security and safeguards is potentially disastrous because the logic is circular; DBT is used to define the security problem in the first place, so it cannot be used to reliably determine the efficacy of the security that was deployed as directed by the DBT model itself.

“Red Teaming” is often held up as a kind of VA, but in the nuclear realm, the term has frequently come to mean a narrowly defined, binary, unrealistic or rigged occasional “test” of a small subset of possible attack scenarios and vulnerabilities.  This “test” is often undertaken by unimaginative personnel burdened by a significant conflict-of-interest in regards to the results.

What is sorely lacking is frequent, independent, imaginative, and comprehensive VAs by personnel who are skilled at finding vulnerabilities and countermeasures, and who are not subject to “shooting the messenger”.  Their findings and recommendations must be objectively evaluated free from organizational wishful thinking and cognitive dissonance.  Good security is proactive, and that requires understanding and managing vulnerabilities in an honest way.  Few nuclear facilities or organizations seem to be able to do this, despite frequent assertions to the contrary.

There is another aspect of VAs that is also underutilized, especially in the nuclear arena.  With something as important as nuclear security and safeguards, there is a natural tendency to want to provide “assurance” that the security and safeguards are adequate.  But this is a difficult and value-judgment kind of problem.  

There is a more effective way to judge security efficacy, which I call the Marginal Technique or (Differential Technique).  The idea is to find the set of security parameters that best minimizes the risk.  This is a complex problem because there are a myriad of possible security parameters, each with a lot of possible values or settings.  Analyzing all possible parameters and settings is not practical.  What makes more sense is to consider our current security parameters and settings, then consider changes to them.  If these changes make the risk go lower, we may want to try more changes of this type.  If the changes make the risk worse, we may want to try other possibilities.   

This is where vulnerability assessors come in. They can tell you how much security would improve if you make any given set of changes.  The goal then is not to seek absolute assurance, but rather to experiment, at least in principle, with security changes to help find ways to reduce vulnerabilities, complicate things for adversaries, and lower overall risk.  We know that we have “pretty good security” when we can’t find any practical changes that significantly decrease our risk.

[For those mathematically inclined (and skip this paragraph if you’re not),  this process is analogous to  plotting the security risk in N-dimensional space against the values of a large number, N-1, of security parameters.  The goal is to find a deep local minimum in this complicated surface in N-dimensional space in order to minimize the risk.  In theory, we would like to find the absolute minimum because this would tell us how to best configure our security.  In practice, however, it is mathematically challenging to find the absolute minimum in a space of many dimensions for such a complex surface.  We should settle instead for a good local minimum and not let the best become the enemy of the good.  In applied mathematics, local minima are often found by taking trial steps in various directions in N-dimensional space in order to determine where the gradient points downward the steepest.  As with any minimization problem in N-dimensional space, it is often wise to consider large changes in order to check whether there are better minimums “over the hill”.  Thus, not all considered changes should be minor tweaks.  Note also that finding the mathematical minimum in N-dimensional space is only an analogy.  Optimizing security is not really a mathematical problem, at least at this stage of our understanding of security.  It is, course, important to bear in mind as well that the risk surface in N-dimensional space is not static, but is always morphing over time with changes in threats, technology, assets we need to protect, resources available to us to provide security, etc.  Thus, we can’t sit permanently at one point in N-dimensional space and think we are good for all time.]

The Marginal Technique can do more than just help us judge our security while avoiding absolutist, binary, wishful-thinking about “assurance”.  With this technique, we are constantly thinking about security and about security changes.  This can help a security program remain flexible, adaptable, and proactive, and avoid the problem of inertia commonly found in large bureaucratic organizations.