
Server Hardening Report: What It Must Contain
A list of remaining gaps is not a server hardening report. The sections a useful report needs, the evidence an auditor can re-check, and the red flags to reject.
The short answer
A useful server hardening report names the baseline used, lists the exact hosts covered, shows what was changed and what was deliberately left alone, proves the changes are in place and the workloads still run, and records every remaining exception with an owner and a review date. A list of remaining gaps or a single benchmark score is a scan result, not a deliverable you can hand to an auditor.
Who this is for
This is for a CTO, head of infrastructure or security lead who has bought, or is about to buy, server hardening and wants to know what the document at the end must prove. It is equally for the person who will show that document to an auditor, a partner's security team or a board. Before the report there is the purchase decision: server hardening services: scan, script or engagement explains which shape of service produces a report worth having. After the report there is the question of staying hardened, which configuration drift after go-live covers. This article is about the middle: the report itself. For the same question on the testing side, see what a real pentest report looks like.
Why a gap list is not a report
Disappointing hardening deliverables fail in the same few ways.
- They describe the benchmark, not your estate. A page of control descriptions copied from a public standard tells you what the standard says. It does not tell you what was true on your hosts on the day.
- They collapse everything into a score. A percentage hides which controls were skipped for a good reason and which were simply missed.
- They list problems without changes. If the provider only reports, you are holding a scan. A hardening engagement is judged by what was changed, validated and handed back.
- They cannot be repeated. Without the baseline in a reusable form and the list of exceptions, nobody can rebuild a host the same way or tell later whether it has drifted.
The report is the part of the engagement a reviewer will read long after the people involved have moved on. That is why its contents belong in the quote, before any host is touched.

The report anatomy matrix
Use this as a checklist when you write the quote or review a draft. It contains no scores or thresholds, because the right targets come from your chosen baseline and your own risk.
| Section | What it must show | Red flag if missing | Who relies on it |
|---|---|---|---|
| Scope and inventory | The hosts, images and environments covered, by name or group, and those excluded | "All servers" with no list | Auditor, system owner |
| Baseline and profile | The named benchmark or internal standard, its version and the profile or level applied | A generic "industry best practice" | Auditor, engineering |
| Method and access | How hosts were assessed and changed, with what access, in which windows | No record of who touched what | Security lead |
| Changes applied | Each change, the control it maps to, the host group, the date and the change reference | A narrative with no control-level detail | Engineering |
| Left alone, with reasons | Controls deliberately not applied, why, and any compensating measure | Silence, which looks like coverage | Everyone |
| Validation | Evidence that workloads still run after the change and that the setting is in place | "Applied successfully" with nothing to check | Product owner, on-call |
| Before and after | The comparison against the baseline, at control level, not only a score | A single percentage | Board, auditor |
| Exceptions register | Remaining gaps, each with an owner, a reason and a review date | Gaps listed with no owner | Risk owner |
| Reusable baseline | The configuration as code or a reproducible build, so new hosts start the same way | A PDF only | Infrastructure team |
| Not covered and next steps | Assets, layers and risks outside the engagement, and the recheck plan | The report implies full coverage | Everyone |
The two rows that separate a real report from a scan export are "left alone, with reasons" and the exceptions register. One shows judgement was exercised; the other shows someone is on the hook for what remains.
Changes applied: the part your team will use
For each group of hosts, a reader should be able to answer the same short list without opening an appendix.
- What changed. The control, the setting before and the setting after, in plain terms. The aim is that an engineer can reverse or repeat it.
- Why it mattered. The risk the control reduces on this kind of host, not a paragraph copied from the benchmark.
- How it was validated. The check that the service still works, such as a health check, a smoke test or a deliberate login, and who ran it.
- What it depends on. Other settings, agents or services that must stay in place for the change to hold.
- Who owns it going forward. The team that now maintains the setting, so a later rebuild does not undo it.
If a section cannot answer these in a few lines, the report describes activity, not results. The same priority logic is explored in hardening versus patching: which your auditor means.
Left alone, with reasons
Every real hardening exercise leaves some controls unapplied: a setting breaks a payment integration, an agent cannot be removed, an older system cannot support a newer option. A trustworthy report says so plainly.
- Reason. The concrete dependency or constraint, not "business requirement".
- Compensating measure. Whatever limits the exposure instead, such as network placement, monitoring or restricted access.
- Owner and review date. A named person and a point at which the decision is revisited.
A report with no exceptions usually means nobody looked hard enough. A report with unexplained exceptions means nobody was asked to defend them. Both fail an auditor's first follow-up question.
Evidence a reviewer can re-check
A report is only as strong as what stands behind it. Ask for these to be named, even if the raw files are held separately.
- Configuration state read from the hosts, collected by you or your team after the change, not only a screenshot from the provider.
- Change references, such as tickets or commits, tying each change to an approval and a window.
- Timestamps in one time zone, so that changes can be lined up against your monitoring and incident history.
- The baseline version, because a result against one version of a benchmark is not a result against another.
- The scan or assessment method used for the before-and-after comparison, so that the second run is comparable to the first.
For a regulated firm, this evidence has to stand up to a reviewer. The DORA baseline for a digital bank's cloud shows how scoped evidence is arranged for a supervisor; the same discipline helps any operator who must answer an auditor.
Before and after: beyond the score
A before-and-after comparison is the section that most readers jump to, and the one most easily flattered. Ask for it at control level, grouped by host type, with the baseline version and the tool used on both runs. A score that rose because controls were removed from scope is not an improvement. A score that stayed flat because exceptions were documented may be an honest result. The comparison should also state which findings were closed by change, which by exception and which remain open.
Red flags when you review a draft
- A report headed by a percentage and nothing underneath it.
- No named baseline, version or profile.
- A host list that says "all servers" or "production".
- Changes described in prose with no control or setting named.
- No exceptions at all, or exceptions with no owner.
- No way for your team to confirm a change on the host.
- The baseline offered only as a PDF, not as code or a reproducible build.
- No recheck plan.
Ask for the report before you sign
Treat the deliverable as a quote line, not an afterthought. Before work starts, ask the provider for a redacted sample report from a past engagement, the headings they will use, the delivery date, whether a walkthrough is included, who keeps the raw data and for how long, and whether a recheck is priced in. If you cannot get a sample, treat that as information. A hardening report also does not replace testing: finding what you did not know is the job of a vulnerability assessment or a penetration test, and a retest question like what counts as fixed applies to hardening findings too.
Limitations
This article is not legal or regulatory advice and does not state what any auditor, supervisor or licensing authority requires of a hardening report. It uses no external statistics and no incident case studies. The matrix is a review checklist to adapt to your own estate, baseline and contract, and it contains no thresholds or scores. A hardening report describes the state of the listed hosts against a named baseline at one point in time; it does not prove that the hosts are free of vulnerabilities, does not cover applications or people, and is not a certification. Configuration drifts after the report is written, and a baseline that suits one workload may break another.
Decision and next step
If you can read the draft and, for each host group, find what changed, why, how it was validated, what was left alone and who owns the exceptions, the report is doing its job. If you cannot, ask for the evidence behind the summary or reject the deliverable. To have the report specified in the quote before any host is changed, get a fixed quote from AnySec and we will map your estate to a baseline, the change windows to agree and the sections you will receive.
Sources and review
The framing follows public guidance: the CIS Benchmarks, which publish configuration baselines with profile levels, and NIST SP 800-123 Guide to General Server Security, which covers securing, maintaining and documenting server configuration, read 2026-11-18; check the current versions, and read your own platform vendors' hardening guidance. AnySec prepared the report matrix from the public structure of hardening engagements; no AnySec-original statistics are used. Author: AnySec Engineering. Published 2026-11-18; reviewed by the AnySec security team.
Related reading
- Server hardening services: scan, script or engagement — which shape of service produces a real report.
- Configuration drift after go-live — keeping the baseline true after the report.
- What a real pentest report looks like — the same deliverable question for penetration testing.
To have the deliverable specified up front, get a fixed quote.
Frequently asked questions
- What should a server hardening report contain?
- The named baseline and profile used, the exact hosts in scope, what was changed and what was deliberately left alone with the reason, evidence that changes were applied and that workloads still run, a before-and-after comparison against the baseline, the exceptions that remain with an owner and a review date, and a statement of what was not covered. A scan export of remaining gaps is not a hardening report.
- Is a benchmark score a good summary of a hardening report?
- Only as one line among others. A single percentage hides which controls were skipped, which were changed and which were accepted as exceptions. Ask for the control-level list behind the score and for the conditions under which a skipped control would be revisited.
- How do we know the changes were actually applied?
- Ask for evidence you can collect yourself: configuration state read from the hosts after the change, the change tickets or commit references, and the timestamps of each change window. A report that only states that controls were applied, without a way for your team to confirm it, is a claim rather than evidence.
- What happens to hardening exceptions?
- Every control left unapplied should be listed as an exception with the reason, the compensating measure if any, a named owner and a review date. Unowned exceptions are how a hardened baseline quietly drifts back to the defaults.
- Do we need a recheck after the report?
- Configuration drifts as hosts are patched, rebuilt and changed. Ask whether a recheck of the same hosts against the same baseline is included, what triggers it, and whether the baseline is delivered as code or a reproducible configuration so new hosts start hardened.
Rather not learn this in production.
Talk to the engineers behind these write-ups — no sales script, a straight read on where you stand.
Get a fixed quote