The mechanics of prioritizing one vulnerability’s business risk over another has always been fraught with concern. What began as securing business applications and infrastructure from full-disclosure bugs a couple of decades ago, has grown to encompass vaguely referenced flaws in insulin-pumps and fly-by-wire aircraft with lives potentially hanging in the balance.
The security industry has always struggled to “score” the significance of the threat posed by a newly discovered vulnerability and recent industry practices have increased pressure on how this should be done.
With the growth of bug bounty programs and vertical industry specialization at boutique security consultancies, vulnerability discoveries with higher severity often translate directly into greater financial reward for the discoverers. As such, there is immense pressure to increase both the significance and perceived threat posed by the vulnerability. In a growing number of cases, marketing teams will conduct world-wide campaigns to alert, scare, and drive business to the company.
It’s been close to 25 years since the first commercial vulnerability scanners started labeling findings in terms of high, medium, and low severity. Even back then, security professionals stumbled by confusing severity with “risk.”
At the turn of the last century as companies battled millennium bugs, the first generation of professional penetration testing consultancies started to include factors such as “exploitability,” “likelihood of exploitation,” and “impact of exploitation” in to their daily reports and end-of-engagement reports as way of differentiating between vulnerabilities with identical severity levels. Customers loved the additional detail, yet the system of scoring was highly dependent on the skills and experience of the consultant tabulating and reporting the results. While the penetration testing practices of 20 years ago have been rebranded Red Teaming and increasingly taken in-house, risk scoring vulnerabilities remains valuable – but continues to be more art than science.
Perhaps the most useful innovation in terms of qualifying the significance of a new vulnerability (or threat) has been the Common Vulnerability Scoring System (CVSS). It’s something I feel lucky to have contributed to and helped drive across products when I led X-Force at Internet Security Systems (acquired by IBM in 2006). As the (then) premier automated scanner and managed vulnerability scanning vendor, the development and inclusion of CVSS v1 scoring back in 2005 changed the industry – and opened up new contentions in the quantitative weighting of vulnerability features that are still wrestled with today in CVSS version 3.1.
CVSS is intended to summarize the severity of vulnerabilities in the context of the software or device – not the systems that are dependent upon the software or device. As a result, it worries me deeply when I hear that CVSS scores are wrongly being used to score the risk a vulnerability poses to an organization, device manufacturer, or end user.
That misconception was captured recently in an article arguing that vulnerability scoring flaws put patients’ lives at risk. On one hand, the researchers point out that though the CVSS score for their newly disclosed vulnerability was only middling (5.8 out of 10), successful exploitation could enable an attacker to adjust medicine dosage levels and potentially kill a patient. And, on the other hand, medical device manufacturers argue that because the score was relatively low, the vulnerability may not require an expedited fix and subsequent regulatory alerting.
As far as CVSS in concerned, both the researchers and medical device vendor were wrong. CVSS isn’t, and should never be used as, a risk score.
Many bright minds over two decades have refined CVSS scoring elements to make it more accurate and useful as a severity indicator, but have stalled in searching for ways to stretch environmental factors and the knock-on impacts of a vulnerability into quantifiable elements for determining “risk.” Today, CVSS doesn’t natively translate to a risk score – and it may never because every industry assesses risk differently and each business has its own risk factor qualifications that an external party won’t know.
I would caution any bug hunter, security analyst, software vendor, or device manufacturer to not rely on CVSS as the pointy end of the stick for prioritizing remediation. It is an important variable in the risk calculation – but it is not an adequate risk qualifier by itself.