Engineering

Which security weaknesses should you fix first?

A long list of software weaknesses gives a manager little guidance. The priority depends on how an attacker could use each one and what would be harmed.

Engineers may bring you a long list of software weaknesses and ask for time to fix them. Some repairs need a service to stop; others require replacing an older system. You need to understand which delays would put customers or operations at risk before agreeing the schedule. Starting with the highest score on the list is not always the right answer. A lower-scoring weakness may matter more if it lets someone change payment details or prevents staff from restarting an important service.

A vulnerability is a weakness that an attacker could use to gain access, disrupt a system or misuse information. Finding one is the start of deciding what to do, not a complete set of instructions for the repair. Ask the team to establish where the affected software is running, how an attacker could use the weakness and what the business would suffer. If the team does not know who runs the system or how it can be reached, assign that investigation. Missing information is not evidence that the problem is harmless.

Check whether the warning applies to the running service

Engineers should compare the installed version and settings with the software supplier’s security notice. Finding an affected component in a stored software bundle is different from finding it running and handling customer requests. The notice may describe particular settings or functions needed for an attacker to use the weakness. Check whether those conditions apply. An incorrect detection is a false positive; a real weakness that is difficult to reach is not.

Connect the finding to the business service and the engineer responsible for it. Include temporary systems, partner connections and older services still processing live information. Looking from outside the company can reveal services accessible over the internet, but not every route through internal systems or supplier access. Engineers need records of what is running and who can access it, alongside the service manager’s knowledge of the work it supports.

Know what the scores can and cannot tell you

A publicly reported vulnerability may have a CVE identifier, such as a reference beginning “CVE” followed by a year and number. CVE stands for Common Vulnerabilities and Exposures. The identifier lets teams check that they are discussing the same weakness; it does not say how urgent the repair is. The Common Vulnerability Scoring System, or CVSS, describes technical severity: how a weakness can be used and the harm it can cause to a system. Its score alone does not establish the likelihood or business cost of an attack on your organisation.

FIRST’s Exploit Prediction Scoring System (EPSS) addresses a different question. It estimates the probability that attackers will use a published CVE somewhere in the real world in the next 30 days. It does not estimate the probability of a successful attack on your company. Keep the date with the prediction so the people making the decision know how current it is.

Evidence that attackers are already using a weakness matters separately. CISA’s Known Exploited Vulnerabilities catalogue lists weaknesses used in real attacks. A listing warrants attention, but does not prove that your system has been attacked. Recent confirmed use should prompt urgent assessment even if EPSS is low. Equally, little public evidence of attacks does not make a weakness safe when it could cause serious damage locally. FIRST’s usage guidance also warns against multiplying CVSS by EPSS. Combining those numbers does not produce a reliable measure of your organisation’s risk.

Ask how an attacker could reach it

A system need not be open to the internet to be reachable by an attacker. Could a partner who is logged in access it? Could someone use stolen login details or start on another internal system and move across? Engineers should describe the route, the permissions required and the protective measures, often called controls, that would stop it. These checks help establish exposure: how accessible the affected software is to attack. Give unresolved questions an investigation owner and deadline rather than treating an incomplete view as permission to wait indefinitely.

Follow what an attacker could do next. Could they stop orders, change information staff rely on, steal customer records or gain access to another system? Check shared login services, backup systems and the tools administrators use to manage other machines. A login service proves who someone is and determines their access. If several systems rely on it, taking it over may affect them all. These identity dependencies also matter during recovery: staff may need that service to sign in before they can restore anything else.

The manager responsible for the service should explain how long work can stop, what customers have been promised and when a change can safely happen. Patching means installing an update that corrects the software. If that needs a scheduled interruption, record when it will happen, what remains at risk until then and how the team will recover if the update fails. Difficulty finding a suitable time is a business problem to resolve, not evidence that the weakness is unimportant.

Distinguish temporary protection from a completed repair

Engineers call a measure that reduces the danger “mitigation”. It may leave the underlying weakness in place. Test that protection in an authorised, controlled environment, recording the software version, settings, attack route and limits of the test. Blocking one attempt proves something about that route, not that every possible attack is blocked. Identify who keeps the protection working, how the team will notice a failure and which system changes require another check.

Remediation means correcting the underlying problem. Before calling it complete, confirm that the corrected software or settings are running on the affected system and that the conditions allowing the attack have been addressed. A warning disappearing from an automated security scan does not, by itself, explain what changed. If an update is impractical, compare limiting use, isolating the system, replacing it or taking it out of service. Review insurance coverage separately; insurance does not repair the weakness or remove the need for someone to manage it.

Agree what would change the repair schedule

Set deadlines using applicable requirements, how the weakness can be reached and what an attack could do to the business. Make sure urgent new evidence of attacks can reach someone authorised to change the schedule. Where a deadline cannot be met, give any permission to delay an owner, conditions and an end date. Reconsider the decision if access or the work the system handles changes. Keep completed repairs, temporary protection and approved delays separate in the work list.

Before approving the plan, ask the engineering team to identify the affected system, the action, the person doing it and the evidence required to close the issue. Review delays caused by missing owners, unchecked repairs and expired permissions as well as the number of open findings. Otherwise the team can close administrative tasks while the business remains vulnerable to the same failure.

Where this applies in our work: the quantification method.