When a hospital group decides to take weakness handling seriously, the first practical argument is rarely about policy. It is about method: how will we actually look at our systems, and how will we decide that something has been dealt with? Each answer has trade-offs in accuracy, disruption and risk. This comparison sets out the main options in a VulnerabilityOps setting, lines them up against the same criteria, and offers a recommendation for organisations that run a mix of office servers, web applications and specialist equipment. The material is administrative and informational, not clinical.
The criteria used here
Four questions are applied to each method in turn. How much of the real picture does it see? How much disturbance can it cause to running systems? What new risk does it introduce? And what evidence does it leave behind for a later review? These match the concerns the Ops Atlas description of VulnerabilityOps raises when it discusses coverage, closure and the handling of scanner credentials.
Part one: how to look
Option A: unauthenticated network scans
An unauthenticated scan probes a machine from the outside without signing in. It is simple to launch, needs no secrets, and shows what a stranger on the network could see. Its weakness is shallow vision. The Atlas notes that such scans miss flaws that only appear locally, inside the operating system or in installed software, so a clean result can be misleading. It suits a perimeter check, such as the booking website or a patient information portal, but it is a poor basis for a closure decision on an internal server.
Option B: authenticated (credentialed) scans
An authenticated scan signs in and inspects the installed software directly. It sees far more, and it is the method the Atlas prefers when it describes verified closure. The costs are real, though. The scanner’s login becomes a powerful secret that must be protected like any privileged account, and the Atlas lists it explicitly among its anti-patterns when treated casually. Logins also expire or break after key rotations, and when they do the report can quietly stop showing findings for the affected machines. For Linux hosts the Atlas adds that a signed-in check also depends on a package inventory being available.
Option C: passive discovery
Passive tools do not probe at all. They watch network traffic and infer what is present. For fragile equipment, this is attractive. The Atlas recommends it for operational technology, where active scanning can destabilise devices. The limitation is that passive observation describes what talks on the network, not what is installed inside, so it supports inventory and exposure tagging much better than it supports proof of a fix.
Option D: image and dependency scanners
For software that the organisation builds or packages itself, scanners that inspect container images and dependency lists, for example the open-source tool Trivy, can run in minutes and leave an exact record of which components were present. They also raise a supply-chain point that the Atlas records: in March 2026 a malicious release of that tool was distributed through redirected tags on its GitHub Action, a reminder that the tools which check your software are themselves software. Pin versions, verify what you pull, and treat the scanner as a component needing its own scrutiny.
Comparing the look-methods side by side
- Depth: authenticated scans and image scanners see the most; unauthenticated scans see the least; passive discovery sees connections but not contents.
- Disturbance: passive discovery is the gentlest, unauthenticated and authenticated scans can load a busy system, and image scanners run away from production altogether.
- New risk: authenticated scanning concentrates privileged logins in one place; image scanners add a dependency on a tool supply chain; the other two add little.
- Evidence value: an authenticated scan or an image digest gives strong closure evidence; the others support discovery and exposure tagging.
Part two: how to decide something is closed
Closure by ticket status
The cheapest method and the weakest. Somebody changes a field to “resolved”. The Atlas treats this as no evidence at all and lists it among the main reasons dashboards look better than reality. It does have one legitimate role, which is as a hand-off record. It must not be the thing that removes an item from the queue.
Closure by rescan
A credentialed rescan that no longer finds the flaw is the standard evidence. Its drawbacks are timing, because it waits for the next scan window, and keying: if the system creates a fresh record each time, the deadline clock restarts. The Atlas suggests that each record carry a key built from three fields: which asset, which flaw, which component, so a later rescan refreshes the existing entry and the deadline continues to run.
Closure by configuration test
Where the remedy is a setting rather than an update, a direct test proves it faster than a full scan. It is precise but narrow, and it only works when someone writes the test. The Atlas pairs it with a reminder from a 2023 appliance flaw: sometimes the definition of fixed must also include ending sessions that were already hijacked, or the attacker simply continues.
Closure by new image digest
For containerised services, a rebuilt image with a different digest and a clean scan is crisp evidence. It suits teams that already build and deploy images routinely, and it does little for appliances or desktop software.
Closure by upheld dispute
Sometimes a finding is wrong for the environment. A reviewed and accepted dispute closes the item. In the Atlas view the team that owns the system should not sign off its own dispute, because that strips out the independent review. A second reviewer, even a part-time one, restores it.
A brief worked comparison
Consider two systems an administrator might face in the same week. The first is a public web form that collects appointment requests. The second is a records archive server that sits on an internal segment and is reachable only by staff. For the web form, an unauthenticated scan gives the outsider’s view and should run frequently; an authenticated scan then confirms the installed components, and the finding closes only on a signed-in rescan. For the archive server, the exposure is lower, so the clock is longer, yet its closure standard is identical. Evidence does not relax because the deadline does.
A third case is a piece of specialised equipment whose supplier controls updates. Here the administrator’s levers are different: passive discovery for visibility, a written agreement with the supplier on dates, and a recorded exception with an expiry when a fix cannot be applied in time. The Atlas counts the total of such exceptions as a debt to be watched, which is a useful way to describe it to management.
Cost and effort follow the same pattern. Unauthenticated scans cost least to operate and are simple to describe to non-specialists. Authenticated scans demand an ongoing credential programme. Image scanning demands a build process that produces images in the first place. Passive discovery demands a place in the network from which traffic can be observed. Budgeting by method, rather than buying one tool and hoping, makes the trade-offs visible before money is spent.
A VulnerabilityOps recommendation for a mixed hospital estate
No single method wins, and the comparison suggests a layered pattern. Use passive discovery for fragile equipment and for building the asset list. Use authenticated scans for internal servers and desktops, with the scanner’s credentials stored and rotated as privileged secrets, and with an alert when coverage drops. Use unauthenticated scans for internet-facing services as a stranger’s-eye view. Use image scanners where applications are packaged in-house. For closure, accept only a rescan, a configuration test, a new digest or an upheld dispute, and keep the ticket status as a hand-off note.
One more precaution follows from the criteria. A database of findings is itself valuable to an attacker, because it lists which machines are weak and how. The Atlas flags it as an attack map, so restrict access to it as you would to any sensitive record.
Where this fits with other work on this site
For neighbouring material, a practical guide to building a safer software journey looks at security within software delivery, while an overview of streamlining software and cloud operations with engineering services covers the operating side. Both complement the method comparison above.
The short version of the verdict: combine looking methods according to what each system can tolerate, and be strict about closure evidence. Organisations that do both can explain, at any point, how a given finding was discovered and why it is considered done.
