Risk, Severity, and Triage

Topics Covered

Why Severity Matters

The cost of everything critical

Vulnerability is not risk

Likelihood

Impact

Context belongs to the owner

What good severity achieves

Severity and priority

Who is harmed

Severity drift

A worked example

Every security programme eventually has more findings than it can fix. Scanners report thousands of vulnerable dependencies, penetration tests list dozens of issues, bug bounty researchers send reports every week, and threat models add their own tickets. If every finding is treated as urgent, engineering teams cannot tell which ones matter, they stop trusting the security team, and the genuinely dangerous issues wait in the same queue as the trivial ones. Severity and triage are the skills that decide what gets fixed this week, what can wait, and what does not need fixing at all.

The cost of everything critical

A common failure looks like this. A dependency scanner is connected to every repository and reports four thousand findings, more than a third of them labelled critical. Tickets are filed automatically. Engineering teams look at the first few, find that the vulnerable function is never called, and conclude that the scanner is noise. The backlog grows, the critical label loses meaning, and when a finding that really is exploitable arrives it is ignored along with the rest. The problem is not the scanner but the absence of a step that separates real risk from theoretical risk before anything reaches an engineer.

With no triage, the backlog grew from 600 to 4,200 findings, 38 percent labelled critical, and the weekly fix rate fell from 40 to 12 as teams stopped trusting it.

Vulnerability is not risk

A vulnerability is a weakness: a flaw in code, configuration or design that could be exploited. Risk is the expected harm from that weakness in a particular context, which depends on how likely exploitation is and how much damage it would cause. The same vulnerable library can be a serious risk in an internet facing service that passes user input to the affected function and almost no risk in a batch job that never calls it. Severity ratings attached to vulnerabilities by their publishers describe the weakness in general. Triage turns them into a judgement about risk in this system.

Likelihood

Likelihood depends on several factors. Reachability asks whether an attacker can get input to the vulnerable code at all, and from where: the internet, an authenticated session, an internal network, or only local access. Exploitability asks how hard the attack is, whether working exploit code exists, and whether attackers are using it now. Exposure asks how many instances are affected and for how long. Existing controls, such as a web application firewall rule, a network restriction or a disabled feature, may reduce likelihood further. A finding with a high base score can have very low likelihood once these are considered.

Impact

Impact asks what an attacker gains if exploitation succeeds. Reading restricted data, changing financial records, taking over accounts, running code on a server with broad cloud credentials and disrupting a service that customers depend on are all high impact. Reading public data, or affecting an isolated test environment, is low impact. Impact also depends on blast radius, the subject of the previous lesson: a vulnerability in a component with narrow privileges has less impact than the same vulnerability in a component that can reach everything. Impact should be described concretely, in terms of data and actions, rather than as an abstract label.

Context belongs to the owner

Much of the context needed to judge risk lives with the team that owns the system, not with the security team. Engineers know whether a function is called, whether input reaches it, which network the service sits on and what data it handles. Security engineers know how attacks work, which vulnerabilities are being exploited and how to read advisories. Good triage combines both, usually through a short structured exchange rather than a long meeting. Asking owners a handful of specific questions, such as is the affected function called with user input, produces better severity decisions than either side could reach alone.

Of 120 findings with critical base scores, 71 were in code never loaded and 40 were not reachable with attacker input. Nine were reachable from the internet.

What good severity achieves

A severity system works when engineers trust it. That trust comes from accuracy: when the security team says something is critical, it is, and it needs attention today. It comes from consistency: two similar findings get similar ratings regardless of who triaged them. It comes from explanation: each rating says why, in a sentence that names the reachability and the impact. And it comes from restraint: the security team is willing to rate things low or close them as not applicable when the evidence supports it. A programme with a few accurate critical findings gets them fixed in days. A programme with hundreds of inflated ones gets none fixed quickly.

Severity and priority

Severity and priority are related but not identical. Severity describes how bad a finding is in context. Priority decides when it is fixed relative to other work, and it can also reflect effort, dependencies and business timing. A medium severity issue that takes ten minutes to fix and sits in code being changed anyway may be fixed before a high severity issue that needs a redesign. Keeping the two separate avoids distorting severity to express urgency, which is how ratings become inflated, and it lets teams make sensible scheduling decisions while still reporting risk honestly.

Who is harmed

A final consideration is who bears the harm. Some findings put the organisation at risk, such as exposure of internal financial data. Others put customers at risk, such as account takeover or disclosure of their personal data. Still others create legal or regulatory exposure, such as breaches that must be reported to authorities. Harm to customers and harm with legal reporting obligations usually deserve higher severity than equivalent harm to internal data, because the organisation cannot contain the consequences on its own. Naming who is harmed in the severity rationale makes this explicit and helps product owners and leaders understand why a finding matters.

Severity drift

Severity is not fixed when it is first assigned. A finding rated low because the affected feature was disabled becomes serious when the feature is turned on. A library vulnerability rated medium because no exploit existed becomes urgent when one is published and attackers begin scanning for it. A service that moves from an internal network to the internet changes the likelihood of every finding in it. Good triage records the assumptions behind each rating so that a change to any of them can trigger a review, and it subscribes to exploitation signals so that ratings rise when the world changes, rather than waiting for the next scheduled review to notice.

A worked example

Consider a single finding: a path traversal flaw in an image processing library, published with a base score of 9.8. In the public upload service, user supplied file names reach the vulnerable function, the service runs with credentials that can read the customer document bucket, and exploit code was published two days ago. The triager rates it critical, explains that any internet user could read arbitrary files including the service credentials, and asks for a fix within three days. In the internal thumbnail job, the same library is loaded but file names are generated by the job itself, so the triager rates it low and schedules the upgrade with the next base image refresh. Both decisions take minutes, and both are written down with their reasons, so anyone reading the tickets later understands why one was urgent and one was not.

Rate the risk in this system, not the weakness in general

A vulnerability becomes risk only in context: how reachable and exploitable it is here and what an attacker would gain. Combine the security team's attack knowledge with the owner's system knowledge, explain each rating in a sentence, keep severity separate from priority, and be willing to rate things low.