About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
Yes, you can usually scan servers you rent, but permission is set by your contract and the provider's testing rules, not by the fact that the servers are yours. A scan sends real traffic through a network that other customers share. Providers therefore write rules about what is allowed, what needs notice, and what is banned outright. Agree the rules before the first packet is sent, in writing, with the provider's named contact.
The second half of the job is reading the results. A scan report is a list of possible weaknesses, not a ranked to-do list. Severity scores, evidence that a flaw is being exploited, and whether the server faces the internet all change the order. This guide is for business owners and IT administrators who want to scan hosted servers or who have just received a report. It uses the testing policies of two large cloud providers as examples of what such rules look like. They are not your provider's rules.
Why it matters
Scanning is part of ordinary patch management. The Australian Signals Directorate's Essential Eight maturity model, as captured for this guide, expects a scanner to be used "at least daily to identify missing patches or updates for vulnerabilities in online services" at its first maturity level, with asset discovery at least fortnightly to find what to scan. If your servers are hosted, someone still has to do this for the layers you are responsible for, and your contract decides which layers those are (our reading: the page is written for organisations and does not address hosted arrangements).
Scanning without agreement carries its own risks. An aggressive scan can slow or crash an application. It can also look like an attack to the provider's monitoring and to other tenants. AWS says that if it receives an abuse report about testing activity it forwards it to the customer, and Microsoft says it "may, at our discretion, interrupt attacks in progress" whether or not they are part of a valid test.
Who may scan what
The first question is whose systems the scan touches. Two parts are easy to confuse:
- Your own servers and applications. Scanning these is generally what testing policies are written to permit, on their terms.
- The provider's infrastructure. This includes the hypervisor, the shared network, management systems and other customers' machines. Policies ban testing these. AWS states that "Customers are not permitted to conduct any security assessments of AWS infrastructure or the AWS services themselves." Microsoft's rules prohibit "testing customer systems without explicit permission".
Providers differ on how much notice they need. AWS says customers can run assessments "without prior approval for the services listed" as permitted, but simulated events such as phishing campaigns or malware tests need a form and, in its words, "must be submitted to AWS at least two (2) weeks in advance of the start date." Microsoft says it encourages "port scanning, or vulnerability assessments on your own Azure Virtual Machines" as long as they are "within your own assets and permissions".
Neither company's policy applies to a smaller or local host. Our reading: the examples show the shape of a policy. Do not assume your provider's terms match them, and do not treat silence as consent.
A related point: AWS also states that "Resellers of AWS services are responsible for their customers' security testing activity." If your host resells another company's infrastructure, there may be two sets of rules, the host's and the upstream provider's (our reading). Ask.
What to agree before a scan
Our recommendation is to agree these points in writing:
- Targets. List the exact IP addresses or names. Exclude anything you cannot show is yours.
- Source addresses. Tell the provider which addresses the scan will come from so its monitoring can tell it from an attack.
- Dates and windows. Choose times when a slow or failed server does the least harm, and avoid backup or batch jobs.
- Intensity and method. State whether the scan is read-only discovery, authenticated checking with credentials, or active testing. Name the tool.
- Banned actions. Both example policies prohibit denial-of-service testing. AWS lists flooding of requests or protocols, and Microsoft lists "network-intensive fuzzing or automated testing that generates excessive traffic". Make sure the scanner's settings cannot cross those lines.
- Contacts and stop conditions. Name who can halt the scan at either end and who is on call.
- Data handling. Agree what happens if the scan finds exposed data. Microsoft's rules tell testers who accidentally reach data they have no rights to to stop, report it and delete it.
- Fixing what you find. Decide in advance who patches which layer.
AWS adds one more principle for tool vendors: tools capable of denial-of-service must be able to disable that function, and the customer is responsible for configuring and validating this. Our reading: check scanner settings before use, not after.
Reading the results
A raw report mixes several kinds of items. A practical order, our recommendation, is below.
1. Confirm the finding is real. Scanners can report a flaw from a version banner without proving it. AWS's policy on denial-of-service (DoS) testing itself describes tools that only perform "a remote query" to learn a software name and version for comparison against versions known to be vulnerable to DoS. Our reading, applying that to scanners generally: such a result is a lead, not proof. For example, a distribution may patch software without changing its version string, so check with the maintainer's advisory or the vendor.
2. Check whether the flaw is being used in attacks. CISA maintains its Known Exploited Vulnerabilities (KEV) catalogue as "the authoritative source of vulnerabilities that have been exploited in the wild" and advises organisations to use it "as an input to their vulnerability management prioritization framework". An item on that list jumps the queue. A second signal is the Exploit Prediction Scoring System (EPSS), a model that "estimates the probability that a published CVE will be exploited in the wild in the next 30 days". Both are inputs, not verdicts.
3. Treat the severity score as a starting point. The Common Vulnerability Scoring System (CVSS) gives a score from 0 to 10. FIRST's specification says base scores are combined "with default values that assume the highest severity for Threat and Environmental metrics", and that consumers "should enrich the Base metrics with Threat and Environmental metric values specific to their use of the vulnerable system". In plain terms, a published score assumes a reasonable worst case. Your server may be more or less exposed.
4. Place it in your environment. Is the server reachable from the internet? Does it hold sensitive data? Is another control already blocking the path? The same flaw on an internet-facing web server and on an internal test machine deserves different urgency (our reading, in line with the Essential Eight's split between internet-facing and other servers).
5. Set a deadline. The Essential Eight captured for this guide sets, at its first maturity level, patching of internet-facing server operating systems "within 48 hours of release when vulnerabilities are assessed as critical by vendors or when working exploits exist", and "within two weeks of release when vulnerabilities are assessed as non-critical by vendors and no working exploits exist". For servers that are not internet-facing, it sets one month at that level, and 48 hours for critical or working-exploit cases only at its third maturity level. These periods run from the vendor's release, not from the day you scan, so a flaw found weeks after its patch appeared may already be past its deadline: say so and treat it as immediate (our reading). Our recommendation: being internal is not a reason to wait on a flaw that is known to be exploited and could cause serious harm; if a patch genuinely cannot go in at once, record the compensating controls (for example, blocking the path to the service) and who accepted the risk. The page displays a last-reviewed date of 27 November 2023, so check the live page for any later version before relying on these numbers.
6. Name an owner. Each finding needs a person and a date, including items the provider must fix and items you must.
A worked example (illustrative)
The following is invented. A firm runs a hosted web server and a hosted database server. It writes to its provider listing both IP addresses, its scanner's source address, a Sunday-morning window and a read-only scan with no denial-of-service checks. The provider acknowledges in writing and gives an on-call number.
The report lists 60 findings. The firm confirms which are real, then finds two on its web server that appear in the CISA catalogue. The firm checks the vendor release dates: one fix came out five days ago, so the 48-hour period for an internet-facing server has already passed, and it asks the provider to apply it immediately; the other came out yesterday, so it asks for the update before 48 hours from that release have passed. The provider manages the operating system, so the firm asks for confirmation in writing. A high-scoring item on the database is not reachable from the internet and is not on the exploited list. The firm checks that and records why it can follow the one-month period, and would move it up straight away if it later appeared on that list. Informational items go on a quarterly list.
Questions to put to your provider
- May we scan our hosted servers, and where is the policy written down?
- What notice, forms or lead time do you need, and who approves?
- What is out of bounds (management network, shared services, other tenants)?
- Which source addresses and tools do you accept, and how do we avoid being blocked?
- Does an upstream provider have its own testing rules?
- Which layers do you patch, which do we, and how do you report patch status?
- Can you provide your own scan or patch evidence so we are not duplicating it?
- Who is the contact if a scan causes a problem?
Read next
Related articles you may want next.
- HostingSeverity levels, response and resolution times: how to read a support scheduleWhat a support response time commits to, how severity is decided, and which questions to ask before you sign a hosting support schedule.Read article
- HostingPatching hosted Windows applications: monthly updates, hotpatch and maintenance windowsHow often a Windows server running business software has to restart for updates, when Microsoft's monthly release lands in Australian time, what hotpatching on Windows Server 2025 really removes, and how to agree a maintenance window the business can live with.Read article
- HostingManaged, co-managed or self-managed hosting: choosing how much of the server you runHow a business decides which layers of a hosted server it hands to a provider and which it keeps, using staff time, the need for administrator access and the evidence it must produce.Read article
Risks and limits
- The AWS and Microsoft pages are examples of other companies' terms. We did not read any local host's policy and make no claim about one.
- Scanning does not prove security. It finds known, detectable issues. It does not find every weakness and may report false positives.
- This is not legal advice; permission also has contractual and legal dimensions we have not covered.
- The Essential Eight timeframes are for the maturity level cited and for the page version captured. The Essential Eight is guidance for organisations and does not by itself set what applies to your hosting arrangement.
- No scanning commands are given. Active scans change network behaviour and are not offered as read-only checks.
Next step
Ask your provider for its scanning permission process in writing, and list the servers and layers you would be responsible for patching. If you are reviewing a hosted service, the Managed Services page is a starting point for that conversation; ask for the scope in writing.
Sources
- Penetration Testing (AWS Customer Support Policy for Penetration Testing), Amazon Web Services, date unavailable.
- Microsoft Security Testing Rules of Engagement, Microsoft Security Response Center, date unavailable.
- CVSS v4.0 Specification Document, version 1.2, FIRST, 18 June 2024.
- Known Exploited Vulnerabilities Catalog, CISA, date unavailable (read via Internet Archive capture of 14 November 2026 because the site refused automated access).
- Exploit Prediction Scoring System (EPSS), FIRST, date unavailable.
- Essential Eight maturity model, Australian Signals Directorate (Cyber.gov.au), last reviewed 27 November 2023 as displayed (read at this current address on 16 November 2026; an Internet Archive capture of 23 October 2026, taken from the page's former address, showed the same text).
All accessed 15 November 2026.
Learning notes
- Your servers are yours to test only on the provider's terms; the provider's own systems are never in scope.
- Agree targets, source addresses, timing, method and contacts in writing before any scan.
- A finding is a lead until confirmed.
- Severity scores assume a worst case; exploitation evidence and exposure change the order.
- Each finding needs an owner and a date, including those the provider must fix.
This article was originally published on .