HIPAA Microsoft 365 Backup: 13 Steps, 90 Min [2026] | #ransomware | #cybercrime

Ransomware crews do not care that your clinic runs on Microsoft 365. In 2025, healthcare providers told Sophos that 34% of attacks ended with data encrypted, and exploited vulnerabilities were the technical root cause in 33% of incidents. If your mailboxes, SharePoint sites and OneDrive folders hold patient information, a HIPAA Microsoft 365 backup is the control that decides whether recovery takes an afternoon or a month.Why Microsoft 365 Needs Its Own HIPAA Backup in AustraliaWhat a Microsoft 365 backup must coverWhat a backup cannot doHIPAA Versus Australian Obligations: What You Actually NeedWhere the data livesPrerequisites and Versions You Need Before Step 1Choose Your Backup Route: Native, Third-Party or BothPhase 1: Scope the Tenant and Harden Backup Admin Access (Steps 1 to 3)Step 1: Inventory what you must protect and price itStep 2: Create a dedicated backup administrator and two break-glass accountsStep 3: Require phishing-resistant MFA for backup adminsPhase 2: Contracts, Billing and Native Protection (Steps 4 to 6)Step 4: Sign the paperwork and fix the billing pathStep 5: Enable Microsoft 365 Backup and create three policiesStep 6: Switch on auditing and keep the evidencePhase 3: Add an Independent, Immutable Copy (Steps 7 and 8)Step 7: Create immutable storage for the second copyStep 8: Shrink the number of people who can destroy thingsPhase 4: Monitoring and Alerting (Steps 9 and 10)Step 9: Define the alerts that matterStep 10: Write down your recovery targets and restore orderPhase 5: Restore Drills and the Ransomware Runbook (Steps 11 to 13)Step 11: Run a mailbox restore drillStep 12: Restore a SharePoint site and a OneDrive accountStep 13: Write the ransomware runbook and the NDB decision treeHow to read your first drill resultsComplete Working Project: A Read-Only Backup Guard for Microsoft 365App registration and layoutgraph.py: authentication and paginginventory.py: size the protected estate and price itguard.py: the daily control checkdrill_log.py: evidence for every restore testRun it and schedule itWhat a HIPAA Microsoft 365 Backup Costs on the Native ServiceCommon Pitfalls That Break Healthcare BackupsTroubleshooting: 12 Problems and FixesAccess and permissionsBackup and restore behaviourAdvanced Tips for Larger Practices and GroupsFAQ: HIPAA Microsoft 365 Backup in AustraliaDoes Microsoft 365 Backup make my clinic HIPAA compliant?Does HIPAA even apply to an Australian healthcare provider?Which Microsoft 365 workloads does the native backup cover?How much does Microsoft 365 Backup cost?Do I still need a third-party backup?Where is my backup data stored?What should I do first if ransomware hits?How often should I test restores?

This tutorial walks an Australian healthcare provider through 13 steps in about 90 minutes of hands-on work. You will pick a backup route, lock down the admin plane, protect Exchange Online, SharePoint Online and OneDrive, wire up monitoring, and rehearse a ransomware restore. It builds on our earlier guides to the HIPAA-compliant cloud backup setup and the HIPAA cloud backup DR plan, but it zooms in on one workload that most clinics run and few back up properly.

One framing point before any clicking. HIPAA is a United States framework. It does not, by itself, make a deployment compliant with Australian privacy or health-record obligations, and Microsoft’s own backup documentation does not claim that it does. Treat HIPAA here as a strong technical benchmark (access control, audit trails, contingency planning, Business Associate Agreements) and layer the Privacy Act and the Notifiable Data Breaches scheme on top.

Google · Preferred Sources
Don’t miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.

Add Now

Why Microsoft 365 Needs Its Own HIPAA Backup in Australia

Many practice managers assume Microsoft 365 is already backed up because the data lives in the cloud. That assumption fails the first time someone deletes a shared mailbox, a compromised admin account purges a SharePoint site, or ransomware syncs encrypted files from a laptop into OneDrive. Retention policies and recycle bins help with accidents. They were not designed as an independent, point-in-time copy that survives a hostile administrator.

The local threat picture keeps getting worse. The Office of the Australian Information Commissioner (OAIC) finalised 1,327 notifications under the Notifiable Data Breaches scheme in 2025-26, an 18% rise and the highest count in the scheme’s history. It also recorded 15 cyber incidents that touched more than one million individuals, up from 2 the year before. Elizabeth Tydd is the Australian Information Commissioner and Carly Kind is the Privacy Commissioner, according to the OAIC’s 2026-27 corporate plan.
You may like

Healthcare providers sit in the middle of that curve. A single compromised receptionist account can expose referral letters, pathology results and Medicare details sitting in a mailbox. A backup that you can restore in stages, with evidence of who did what and when, turns a notifiable breach into a contained one. It does not replace prevention, but it caps the damage.

Patient trust is part of the business case. The OAIC published the key findings of its 2026 Australian Community Attitudes to Privacy Survey, and the document is worth reading before you brief your partners or board. A clinic that can show a tested restore plan has a far easier conversation with patients after an incident than one that cannot.

What a Microsoft 365 backup must cover

At minimum, the protected scope should include Exchange Online mailboxes (including archives), SharePoint Online sites, and OneDrive for Business accounts. Many clinics also depend on Teams chats and channel files, and that is where a gap appears. As of September 30, 2026, Teams backup in Microsoft 365 Backup is not released and remains listed as coming in the future. If Teams carries clinical conversations, you need a third-party tool or a documented compensating control.

What a backup cannot do

Windows 11 now backs up device settings by default on eligible machines. That is a convenience feature for settings, not a tenant-level copy of patient data, so do not count it. Likewise, consumer OneDrive storage changes announced by Microsoft in October 2026 have nothing to do with business backup for a healthcare tenant. Keep your scope clear and your auditor will thank you.

HIPAA Versus Australian Obligations: What You Actually Need

The two regimes overlap on engineering controls and diverge on legal mechanics. HIPAA’s Security Rule expects a data backup plan, a disaster recovery plan, access controls, audit controls and transmission security. Australian providers face the Privacy Act, the Australian Privacy Principles (particularly APP 11 on security of personal information), the NDB scheme and, for My Health Record participants, additional rules. Use the table below as a working map, not as legal advice.

Control areaHIPAA expectationAustralian equivalent to checkWhere this tutorial covers itBackup and recovery planContingency plan with data backup and disaster recoveryAPP 11 reasonable steps to protect personal informationSteps 4, 11, 12Access controlUnique user IDs, least privilegeAPP 11 and internal access policiesSteps 2, 3Audit trailAudit controls on systems with patient dataEvidence for OAIC assessment of a suspected breachSteps 6, 9Third-party processorsBusiness Associate AgreementContract terms and overseas disclosure checks (APP 8)Step 4Breach responseBreach notification ruleNDB scheme assessment and notificationStep 13TestingPeriodic evaluation of the planEvidence that safeguards workSteps 11, 12

The practical upshot: build to the stricter of the two on every row. If you also hold a BAA with a US-linked vendor, read our walkthrough on how to sign a HIPAA BAA for cloud backup before you enable anything that stores copies outside Australia.

Where the data lives

Data location is the question that trips up most Australian clinics. Microsoft’s certification directory lists the storage and processing regions each backup application uses. The Veritas NetBackup SaaS Protection entry, for example, names Australia plus several other regions including the United States, Canada, Ireland, Japan, Singapore and the United Kingdom. Alcion’s entry reportedly lists the United States, Australia, Ireland, Germany and Singapore. A listing that includes Australia does not guarantee your copies stay there, so confirm the region setting in writing with the vendor.
Prerequisites and Versions You Need Before Step 1

Set aside 90 minutes and a quiet change window. Nothing in the first nine steps interrupts clinical work, but Steps 11 and 12 restore real objects into a test location, so tell your practice manager before you start. Gather the following.

A Microsoft 365 tenant with Exchange Online, SharePoint Online and OneDrive in use, plus Global Administrator access for the setup session (you will drop it afterwards).An Azure subscription you control. Microsoft 365 Backup bills pay-as-you-go through Azure, and the documented model has no separate license or upfront commitment.Two FIDO2 security keys per backup administrator (one primary, one stored in a safe).A current version of PowerShell, plus the latest Microsoft Graph and Exchange Online modules.A current version of Python for the monitoring project.A signed contract and data-processing terms from any third-party backup vendor, including written confirmation of storage regions.A test SharePoint site and a test mailbox that contain no real patient data, used only for restore drills.

PhaseStepsTime budgetWho should be in the room1. Scope and harden1 to 320 minutesIT lead, practice manager2. Contracts and native protection4 to 620 minutesIT lead, finance, privacy officer3. Independent immutable copy7 and 810 minutesIT lead, cloud engineer4. Monitoring and targets9 and 1015 minutesIT lead, clinical lead5. Drills and runbook11 to 1325 minutes plus waiting timeIT lead, privacy officer, practice owner

Restore drills include waiting time while data copies, so the clock time for Phase 5 stretches beyond the hands-on budget. Book the privacy officer early. Their sign-off on Step 13 is the piece most often delayed, and a runbook without that sign-off is a draft, not a control.

Stay current on tooling, because software that talks to your tenant changes often. MSP360, for instance, released Backup for Microsoft 365 version 4.11 on October 20, 2026 and, per its release notes, moved from Exchange Web Services to the Microsoft Graph API. Any tool still on EWS deserves a hard look from your security team.

Choose Your Backup Route: Native, Third-Party or Both

You have three realistic routes. First, Microsoft 365 Backup, the first-party service managed from the Microsoft 365 admin center. It protects Exchange Online, SharePoint Online and OneDrive for Business, and Microsoft lists it at US$0.15 per GB per month of protected content. Second, a certified third-party product that sits on Microsoft 365 Backup Storage or uses its own storage. Third, a hybrid in which the native service handles fast restores and an independent vendor holds a separate copy in a different trust domain.

Third-party options in Microsoft’s app certification directory reportedly include Alcion, Veritas NetBackup SaaS Protection and AvePoint Cloud Backup for Microsoft 365, with records updated in the second half of October 2026. The Alcion listing reportedly answers yes to HIPAA compliance, but the listing says nothing about Australian law and does not make your deployment compliant. Veritas documentation, now under Cohesity branding, still describes NetBackup SaaS Protection hubs, so the product is not retired. We could not confirm current pricing for these three, so request quotes in writing.

If you want a worked example of a specialist vendor, our guides to Keepit Microsoft 365 backup and Veeam Backup for Microsoft 365 cover two other routes in full. The steps below stay vendor-neutral where they can, and name the native service where Microsoft’s behaviour matters.

Question to ask the vendorWhy it matters for a clinicEvidence to requestWhich workloads are covered?Teams backup in the native service is not released, so gaps are realWritten workload list with datesWhere are copies stored and processed?APP 8 overseas disclosure and HIPAA BAA scopeRegion list in the contractWill you sign a BAA?Required if a US-regulated entity or HIPAA benchmark appliesExecuted agreementAre backups immutable?A hostile admin must not be able to purge copiesTechnical description and a live testWho can delete backup data?Separation of duties between tenant admin and backup adminRole matrixHow do offboarding charges work?Native restore points keep billing until they expirePricing page and a worked exampleWhat audit logs can I export?Evidence for an NDB assessmentSample log export

Our recommendation for most clinics under 100 staff: run the native service for fast restores and add one independent copy if your risk assessment demands separation from the Microsoft tenant. The sections below assume that hybrid shape but work equally if you pick only one route.

Phase 1: Scope the Tenant and Harden Backup Admin Access (Steps 1 to 3)

Step 1: Inventory what you must protect and price it

You cannot size a backup, or its bill, without counting mailboxes, sites and OneDrive accounts. The Graph usage reports give you this in a minute. Sign in with a read-only scope and pull 30-day reports.

Connect-MgGraph -Scopes "Reports.Read.All" -NoWelcome

New-Item -ItemType Directory -Force -Path ./inventory | Out-Null
Get-MgReportMailboxUsageDetail -Period D30 -OutFile ./inventory/mailboxes.csv
Get-MgReportSharePointSiteUsageDetail -Period D30 -OutFile ./inventory/sites.csv
Get-MgReportOneDriveUsageAccountDetail -Period D30 -OutFile ./inventory/onedrive.csv

foreach ($f in "mailboxes","sites","onedrive") {
$rows = Import-Csv "./inventory/$f.csv"
$bytes = ($rows | Measure-Object -Property "Storage Used (Byte)" -Sum).Sum
"{0,-10} count={1,5} GB={2,9:N1}" -f $f, $rows.Count, ($bytes / 1GB)
}

The output looks like this for a mid-sized practice (illustrative figures from a lab tenant, not a benchmark):

mailboxes count= 64 GB= 412.8
sites count= 23 GB= 236.4
onedrive count= 58 GB= 497.1

Estimate your own tenant by multiplying the protected gigabytes by the list price of US$0.15 per GB per month. Remember that billing counts recycle-bin content and Exchange online archives too. Microsoft’s own example charges 3.5 GB for a 1 GB site holding 0.5 GB in its second-stage recycle bin plus a 1 GB mailbox with a 1 GB archive. Add 10 to 15 percent headroom when you budget.

Step 2: Create a dedicated backup administrator and two break-glass accounts

Separation of duties is the HIPAA concept that pays off most under ransomware. The person who can delete a mailbox should not be the same identity that can delete its backup. Create a cloud-only account such as [email protected], assign only the roles Microsoft’s current Backup documentation lists for administering the service, and never use it for email or browsing.

Then create two break-glass accounts. Give each a long random passphrase stored in separate sealed envelopes, register a FIDO2 key on each, and exclude them from your Conditional Access policies so a policy mistake cannot lock you out. Alert on every sign-in from these accounts in Step 9.

Step 3: Require phishing-resistant MFA for backup admins

Start in report-only mode so you can see who would be blocked before you enforce anything. This policy targets a group named for backup admins and uses the built-in Phishing-resistant MFA authentication strength.

Connect-MgGraph -Scopes "Policy.ReadWrite.ConditionalAccess","Group.Read.All" -NoWelcome

$grp = (Get-MgGroup -Filter "displayName eq 'SG-Backup-Admins'").Id
$bg1 = (Get-MgUser -UserId "[email protected]").Id
$bg2 = (Get-MgUser -UserId "[email protected]").Id

$policy = @{
displayName = "HIPAA-BKP: phishing-resistant MFA for backup admins"
state = "enabledForReportingButNotEnforced"
conditions = @{
users = @{ includeGroups = @($grp); excludeUsers = @($bg1, $bg2) }
applications = @{ includeApplications = @("All") }
}
grantControls = @{
operator = "OR"
authenticationStrength = @{ id = "00000000-0000-0000-0000-000000000004" }
}
}
New-MgIdentityConditionalAccessPolicy -BodyParameter $policy | Select-Object Id, DisplayName, State

Watch the sign-in logs for a week, confirm that every admin has a registered key, then flip the state to enabled. If you only do one thing from this article this week, do this one. Stolen sessions and phished passwords remain the quickest way for an attacker to delete a backup.
Phase 2: Contracts, Billing and Native Protection (Steps 4 to 6)

Step 4: Sign the paperwork and fix the billing path

Technical controls mean little without contracts that match them. For every vendor that touches patient data, collect three documents: the data-processing terms, a BAA if you are working to the HIPAA benchmark, and a region statement naming where copies and metadata live. Our HIPAA vendor selection guide includes a due-diligence questionnaire you can reuse here.

Then sort out billing. The native service bills through an Azure subscription, so create a dedicated resource group and a budget alert before you switch anything on. Microsoft has moved pay-as-you-go billing management into the Billing experience of the Microsoft 365 admin center for customers onboarded earlier, so older screenshots in blog posts may not match what you see. Customers who buy through an independent software vendor are billed by that vendor instead, with no separate Microsoft charge for backup storage.

Step 5: Enable Microsoft 365 Backup and create three policies

Open the Microsoft 365 admin center as the dedicated backup administrator and search for Microsoft 365 Backup. Microsoft moves menu entries around, so searching beats following a remembered path. Link the Azure subscription and resource group from Step 4, then create one policy per workload so you can restore each independently.

Exchange policy: add clinical and administrative mailboxes, shared mailboxes for referrals and results, and any mailbox with an online archive.SharePoint policy: add sites holding clinical policies, rosters, referral templates and scanned documents.OneDrive policy: add accounts of clinicians and administrators who store working files.

Keep a spreadsheet of what each policy includes and who approved it. When a nurse joins or leaves, the policy membership must change the same day. A mailbox that nobody added to the policy is the most common reason a restore fails when you need it.
You may like

Remember the offboarding rule before you experiment. Turning the service off starts the offboarding process, but removing a workload from a policy does not necessarily stop charges at once because retained restore points keep contributing to billable usage until they expire. Do not create test policies against production data and walk away.

Step 6: Switch on auditing and keep the evidence

HIPAA’s audit controls and an OAIC assessment both depend on logs that exist before the incident. Confirm that unified audit log ingestion and mailbox auditing are on, and note the retention your licence provides.

Connect-ExchangeOnline -ShowBanner:$false

Get-AdminAuditLogConfig | Select-Object UnifiedAuditLogIngestionEnabled
Get-OrganizationConfig | Select-Object AuditDisabled

# Turn ingestion on if it reports False
Set-AdminAuditLogConfig -UnifiedAuditLogIngestionEnabled $true

UnifiedAuditLogIngestionEnabled
-------------------------------
True

AuditDisabled
-------------
False

Export audit records to storage the tenant administrators cannot alter. Our guide to auditing a HIPAA cloud backup explains the two-year log pattern and how to prove it to an assessor.

Phase 3: Add an Independent, Immutable Copy (Steps 7 and 8)

Step 7: Create immutable storage for the second copy

If a single compromised tenant admin can delete both your data and your backup, you have one copy, not two. A second copy in a different trust domain, with object lock and keys you control, closes that gap. The example below creates a Sydney-region bucket with Object Lock for a vendor or script that supports bring-your-own storage. Confirm that your chosen product supports this before you build it.

aws s3api create-bucket \
--bucket yourclinic-m365-vault \
--region ap-southeast-2 \
--create-bucket-configuration LocationConstraint=ap-southeast-2 \
--object-lock-enabled-for-bucket

aws s3api put-object-lock-configuration \
--bucket yourclinic-m365-vault \
--object-lock-configuration '{"ObjectLockEnabled":"Enabled","Rule":{"DefaultRetention":{"Mode":"COMPLIANCE","Days":35}}}'

aws s3api put-bucket-encryption \
--bucket yourclinic-m365-vault \
--server-side-encryption-configuration '{"Rules":[{"ApplyServerSideEncryptionByDefault":{"SSEAlgorithm":"aws:kms","KMSMasterKeyID":"alias/clinic-m365-vault"},"BucketKeyEnabled":true}]}'

Compliance mode means nobody, including the account root user, can shorten retention or delete locked versions until the period ends. That is exactly what you want against ransomware and exactly what you do not want if you mistakenly load the wrong data. Test in a throwaway bucket first. For key design, see our guide to customer-managed keys for HIPAA backups, and for retention periods read the HIPAA backup retention walkthrough.

Step 8: Shrink the number of people who can destroy things

Count your Global Administrators. Most small clinics discover six or seven, including a former IT contractor. Every extra one is another phishing target who can purge data.

Connect-MgGraph -Scopes "RoleManagement.Read.Directory","Directory.Read.All" -NoWelcome

$role = Get-MgDirectoryRole -Filter "displayName eq 'Global Administrator'"
Get-MgDirectoryRoleMember -DirectoryRoleId $role.Id |
ForEach-Object { $_.AdditionalProperties.userPrincipalName } | Sort-Object

Aim for fewer than five, with two of them the break-glass accounts. Move daily admin work to scoped roles, and require approval for elevation if your licence includes privileged identity management. For a broader control map that includes backups, our Essential Eight M365 and Azure backup guide shows how the same steps feed the maturity model.
Phase 4: Monitoring and Alerting (Steps 9 and 10)

Step 9: Define the alerts that matter

A backup nobody watches fails silently. Managed backup providers in Australia advertise continuous monitoring, integrity checks and monthly status reports, and you should demand the same of your own setup. Start with the seven alerts in the table below. Route them to a monitored shared mailbox and a phone, not to the inbox of the person who just left.

AlertSignal sourceTriggerFirst responseBackup protection disabledAdmin center and audit logPolicy changed or workload removedCall the policy owner and verify intentMass deletionUnified audit logDelete events far above your daily baselineSuspend the account, start the runbookBreak-glass sign-inEntra sign-in logsAny sign-inConfirm with both custodiansNew Global AdministratorEntra audit logRole assignmentApprove or revert within the hourFailed authentication burstEntra sign-in logsRepeated failures on backup adminReset keys, review Conditional AccessStorage anomalyAzure Cost ManagementDaily spend above budgetCheck for new workloads or attack stagingRestore drill overdueYour drill logNo successful drill in 90 daysSchedule a drill this week

Pick thresholds from your own baseline. A clinic that deletes ten items a day should alert at a few hundred, while a records department that bulk-archives each quarter needs a calendar exception. Tune for two weeks, then freeze the numbers and record them in your policy document.

Step 10: Write down your recovery targets and restore order

Do not accept a vendor’s claim that most workloads restore within hours. Treat it as a hypothesis and measure it in Steps 11 and 12. Before you test, agree targets with the clinical lead. A sensible starting order for a clinic looks like this:

Identity first: confirm Entra ID is clean, reset credentials and revoke sessions.Exchange Online: shared mailboxes for referrals and results, then clinician mailboxes.SharePoint Online: clinical policies, rosters and templates.OneDrive for Business: working files, restored per user as people return to work.Teams content: handled through your compensating control, since native Teams backup is not released.

Write an RPO and RTO beside each item, signed by the practice owner. Turn those numbers into a quarterly test schedule that an assessor will accept, and keep each drill record next to the targets it proves.

Phase 5: Restore Drills and the Ransomware Runbook (Steps 11 to 13)

Step 11: Run a mailbox restore drill

Use the test mailbox from the prerequisites. Seed it with 50 dummy messages, wait for a backup cycle, delete 20, then restore to the point before deletion. Time the whole sequence with a stopwatch and record it. Restore into a new location first so you never overwrite live data during a drill.

python3 drill_log.py start --item "Exchange: test mailbox" --expected-items 50
# ... perform the restore in the admin center or vendor console ...
python3 drill_log.py finish --item "Exchange: test mailbox" --restored-items 50 --notes "restored to new mailbox folder"

[2026-10-25T10:12:04+11:00] START Exchange: test mailbox expected=50
[2026-10-25T10:31:47+11:00] FINISH Exchange: test mailbox restored=50 elapsed=19m43s result=PASS

Those timings come from a lab run and will differ in your tenant. What matters is that the result lands in a log with a date, a person and a pass or fail. The script that produces it is in the project section below.

Step 12: Restore a SharePoint site and a OneDrive account

Repeat the drill for a test SharePoint site and a test OneDrive account. Delete a folder tree, restore it, then compare file counts and a few checksums against a manifest you created before the deletion. Counting files is not enough. Open three documents and confirm that permissions, version history and sharing links look right, since a restore that returns bytes but loses access control is a different kind of incident.
Read nextMicrosoft MAI Models: 7 In-House AIs Rival OpenAI [2026]AI & Machine Learning · Jun 26

Also test the independent copy from Step 7. Restoring from a different trust domain exercises a different path with different credentials, and that is the path you will use if the tenant itself is compromised. If it needs a person who is on leave, you have found a staffing gap worth fixing now.

Step 13: Write the ransomware runbook and the NDB decision tree

Print this and keep a copy offline. In an incident nobody wants to read documentation on a dead laptop.

Isolate. Disable the suspected accounts, revoke refresh tokens, and block sync clients so encrypted files stop flowing into OneDrive and SharePoint.Preserve evidence. Export audit and sign-in logs before anyone changes settings.Freeze backup changes. No one edits backup policies or deletes restore points during the incident.Find the last clean point. Compare audit timestamps of the first malicious activity with available restore points and pick one before it.Restore in stages. Follow the order from Step 10 and verify each stage before the next.Assess notification. Decide whether an eligible data breach occurred under the NDB scheme, based on whether personal information was accessed or released and whether serious harm is likely. Record your reasoning and the date.Review. Within two weeks, hold a lessons-learned session and update the runbook.

A backup helps with restoration and investigation, but it does not satisfy the NDB scheme by itself. The OAIC says it finalised 88% of 2025-26 notifications within 60 days, against a target of 80%, which shows the regulator expects well-documented assessments. Keep your notes tight and dated.

How to read your first drill results

Expect the first drill to expose problems. That is the point. A pass on item count with a failure on permissions tells you the restore path works but the access model needs a fix. A slow restore tells you your RTO is fiction and needs either a bigger window or a different tool. A restore that needed an engineer who was on leave tells you to document the procedure and train a second person.

Log each finding as an action with an owner and a due date, then rerun the same drill after the fix. Assessors care less about a perfect first result than about a visible loop of test, fix and retest. Keep the CSV from drill_log.py alongside the corrective-action list so one folder tells the whole story.
Complete Working Project: A Read-Only Backup Guard for Microsoft 365

The steps above leave you with settings. This project adds the habit: a small Python toolkit that inventories your tenant, checks the controls from Steps 3, 8 and 9 every morning, and logs restore drills. It uses read-only Microsoft Graph permissions, so a bug in the script cannot change your tenant.

App registration and layout

In Entra ID, register an application named m365-backup-guard. Add these application permissions on Microsoft Graph and grant admin consent: Reports.Read.All, Directory.Read.All, AuditLog.Read.All and Policy.Read.All. Use a certificate or a short-lived secret kept in Azure Key Vault. Never paste the secret into a script. The project layout is:

m365-backup-guard/
requirements.txt
config.json
graph.py
inventory.py
guard.py
drill_log.py

# requirements.txt
msal
requests

{
"price_per_gb_month_usd": 0.15,
"headroom_pct": 15,
"max_global_admins": 5,
"breakglass_upns": ["[email protected]", "[email protected]"],
"required_ca_policy_prefix": "HIPAA-BKP:",
"max_days_since_drill": 90,
"drill_log": "drills.csv"
}

graph.py: authentication and paging

import os
import msal
import requests

GRAPH = "https://graph.microsoft.com/v1.0"

def _token():
app = msal.ConfidentialClientApplication(
os.environ["GUARD_CLIENT_ID"],
authority=f"https://login.microsoftonline.com/{os.environ['GUARD_TENANT_ID']}",
client_credential=os.environ["GUARD_CLIENT_SECRET"],
)
result = app.acquire_token_for_client(scopes=["https://graph.microsoft.com/.default"])
if "access_token" not in result:
raise SystemExit(f"Token error: {result.get('error_description')}")
return result["access_token"]

_headers = None

def headers():
global _headers
if _headers is None:
_headers = {"Authorization": f"Bearer {_token()}"}
return _headers

def get(path, params=None):
r = requests.get(f"{GRAPH}{path}", headers=headers(), params=params, timeout=60)
r.raise_for_status()
return r

def get_all(path, params=None):
url, items = f"{GRAPH}{path}", []
while url:
r = requests.get(url, headers=headers(), params=params, timeout=60)
r.raise_for_status()
body = r.json()
items.extend(body.get("value", []))
url, params = body.get("@odata.nextLink"), None
return items

inventory.py: size the protected estate and price it

import csv
import io
import json
from graph import get

REPORTS = {
"mailboxes": "/reports/getMailboxUsageDetail(period='D30')",
"sites": "/reports/getSharePointSiteUsageDetail(period='D30')",
"onedrive": "/reports/getOneDriveUsageAccountDetail(period='D30')",
}

def load(path):
text = get(path).content.decode("utf-8-sig")
return list(csv.DictReader(io.StringIO(text)))

def main():
cfg = json.load(open("config.json"))
total_gb = 0.0
print(f"{'workload':<10} {'count':>6} {'GB':>10}")
for name, path in REPORTS.items():
rows = load(path)
gb = sum(float(r.get("Storage Used (Byte)") or 0) for r in rows) / 1024**3
total_gb += gb
print(f"{name:<10} {len(rows):>6} {gb:>10.1f}")
padded = total_gb * (1 + cfg["headroom_pct"] / 100)
cost = padded * cfg["price_per_gb_month_usd"]
print(f"\nTotal {total_gb:.1f} GB, with headroom {padded:.1f} GB")
print(f"List-price estimate: US${cost:,.2f} per month at US${cfg['price_per_gb_month_usd']}/GB")

if __name__ == "__main__":
main()

guard.py: the daily control check

import csv
import json
import sys
from datetime import datetime, timedelta, timezone
from graph import get, get_all

def check_global_admins(cfg):
role = get("/directoryRoles", {"$filter": "displayName eq 'Global Administrator'"}).json()["value"][0]
members = get_all(f"/directoryRoles/{role['id']}/members")
n = len(members)
return n <= cfg["max_global_admins"], f"{n} Global Administrators (limit {cfg['max_global_admins']})"

def check_ca_policies(cfg):
pols = get_all("/identity/conditionalAccess/policies")
mine = [p for p in pols if p["displayName"].startswith(cfg["required_ca_policy_prefix"])]
enforced = [p for p in mine if p["state"] == "enabled"]
return bool(enforced), f"{len(enforced)} enforced of {len(mine)} backup policies found"

def check_breakglass(cfg):
since = (datetime.now(timezone.utc) - timedelta(hours=24)).strftime("%Y-%m-%dT%H:%M:%SZ")
hits = []
for upn in cfg["breakglass_upns"]:
flt = f"userPrincipalName eq '{upn}' and createdDateTime ge {since}"
hits += get_all("/auditLogs/signIns", {"$filter": flt})
return not hits, f"{len(hits)} break-glass sign-ins in the last 24 hours"

def check_drill(cfg):
try:
rows = list(csv.DictReader(open(cfg["drill_log"])))
except FileNotFoundError:
return False, "no drill log found"
passes = [r for r in rows if r["result"] == "PASS"]
if not passes:
return False, "no passing drill recorded"
last = datetime.fromisoformat(max(r["finished"] for r in passes))
age = (datetime.now(last.tzinfo) - last).days
return age <= cfg["max_days_since_drill"], f"last passing drill {age} days ago"

def main():
cfg = json.load(open("config.json"))
checks = [
("global-admin-count", check_global_admins),
("ca-policy-enforced", check_ca_policies),
("breakglass-quiet", check_breakglass),
("restore-drill-fresh", check_drill),
]
failed = 0
for name, fn in checks:
ok, detail = fn(cfg)
failed += 0 if ok else 1
print(f"{'OK ' if ok else 'FAIL'} {name:<22} {detail}")
sys.exit(1 if failed else 0)

if __name__ == "__main__":
main()

drill_log.py: evidence for every restore test

import argparse
import csv
import json
import os
from datetime import datetime

def now():
return datetime.now().astimezone().replace(microsecond=0)

def main():
ap = argparse.ArgumentParser()
ap.add_argument("action", choices=["start", "finish"])
ap.add_argument("--item", required=True)
ap.add_argument("--expected-items", type=int, default=0)
ap.add_argument("--restored-items", type=int, default=0)
ap.add_argument("--notes", default="")
a = ap.parse_args()

cfg = json.load(open("config.json"))
state = f".drill-{a.item.replace(' ', '_').replace(':', '')}.json"

if a.action == "start":
json.dump({"started": now().isoformat(), "expected": a.expected_items}, open(state, "w"))
print(f"[{now().isoformat()}] START {a.item} expected={a.expected_items}")
return

s = json.load(open(state))
started, finished = datetime.fromisoformat(s["started"]), now()
secs = int((finished - started).total_seconds())
result = "PASS" if a.restored_items >= s["expected"] else "FAIL"
new = not os.path.exists(cfg["drill_log"])
with open(cfg["drill_log"], "a", newline="") as f:
w = csv.writer(f)
if new:
w.writerow(["item", "started", "finished", "seconds", "expected", "restored", "result", "notes"])
w.writerow([a.item, started.isoformat(), finished.isoformat(), secs, s["expected"], a.restored_items, result, a.notes])
os.remove(state)
print(f"[{finished.isoformat()}] FINISH {a.item} restored={a.restored_items} elapsed={secs//60}m{secs%60:02d}s result={result}")

if __name__ == "__main__":
main()

Run it and schedule it

python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt

export GUARD_TENANT_ID="your-tenant-guid"
export GUARD_CLIENT_ID="your-app-client-id"
export GUARD_CLIENT_SECRET="$(az keyvault secret show --vault-name clinic-kv --name guard-secret --query value -o tsv)"

python3 inventory.py
python3 guard.py

workload count GB
mailboxes 64 412.8
sites 23 236.4
onedrive 58 497.1

Total 1146.3 GB, with headroom 1318.2 GB
List-price estimate: US$197.73 per month at US$0.15/GB

OK global-admin-count 4 Global Administrators (limit 5)
OK ca-policy-enforced 1 enforced of 1 backup policies found
OK breakglass-quiet 0 break-glass sign-ins in the last 24 hours
FAIL restore-drill-fresh no drill log found

That last FAIL is deliberate on day one. It turns green after your first logged drill. Schedule guard.py from cron, Azure Automation or a GitHub Actions workflow with the secret held in the platform’s vault, and send a non-zero exit code to your alert channel. Pair it with an independent, ransomware-resistant copy such as the one in our Arq ransomware-proof backup tutorial for the endpoints that sit outside Microsoft 365.
What a HIPAA Microsoft 365 Backup Costs on the Native Service

Microsoft lists the first-party service at US$0.15 per GB per month of protected content, pay-as-you-go, with no upfront commitment. The Microsoft Learn pricing page was last updated October 24, 2026, so check it before you present numbers to your board. The table converts that list price into monthly figures, with and without the 15 percent headroom we recommended in Step 1. Your invoice will show the exact currency and any tax.

Protected contentMonthly at list priceWith 15% headroomAnnual with headroom500 GBUS$75.00Not listedNot listed1,000 GBNot listedNot listedNot listed2,000 GBUS$300.00Not listedNot listed5,000 GBUS$750.00Not listedNot listed

Add the cost of any independent second copy. That figure depends on the vendor or the object storage you pick, and we could not confirm current list prices for the third-party products named above. For wider context on infrastructure costs, browse the cloud computing hub before you negotiate.

Common Pitfalls That Break Healthcare Backups

Treating retention as backup. Recycle bins and retention policies live in the same tenant an attacker will control. They help with accidents and nothing else.One admin to rule both worlds. If the identity that can delete a mailbox can also delete its restore points, ransomware needs one stolen session, not two.Assuming HIPAA equals Australian compliance. A vendor ticking the HIPAA box in a directory says nothing about the Privacy Act, APP 8 or your contract.Forgetting Teams. Clinicians use chat for urgent handovers. If native Teams backup is not released, write down your compensating control or accept the gap in your risk register.Letting sync clients spread damage. Encrypted files on a laptop sync upward within minutes. Your runbook must block sync early.Skipping restore drills. Untested backups are guesses. Schedule drills quarterly and log them.Ignoring offboarding charges. Removing a workload from a policy does not end billing until retained restore points expire.Stale policy membership. New mailboxes and sites do not protect themselves. Tie policy updates to your joiner and leaver process.

Troubleshooting: 12 Problems and Fixes

Access and permissions

Graph reports return 403. The app lacks Reports.Read.All or admin consent was never granted. Add the permission, click Grant admin consent, and wait a few minutes for the token cache to refresh.Report CSVs show hashed user names. A tenant setting conceals names in usage reports. An admin can turn concealment off under the reports settings in the Microsoft 365 admin center, or you can work with counts only.Sign-in log queries fail with an authorisation error. The sign-in log API needs AuditLog.Read.All and a licence tier that includes Entra sign-in logs. Check both before debugging the filter.A Conditional Access policy locks out an admin. Use a break-glass account, switch the policy back to report-only, then register a key for the affected admin. This is why Step 3 starts in report-only mode.Microsoft Graph PowerShell throws assembly errors. Multiple SDK versions are loaded. Close all sessions, update the modules to the latest version, and start a clean PowerShell 7 window.

Backup and restore behaviour

The bill is higher than your estimate. Billing counts recycle-bin content and Exchange online archives. Re-run the inventory with those in mind and add headroom.A restore returns fewer items than expected. The item may predate the restore point, or the mailbox or site was never in a policy. Check policy membership first.A restore overwrote live content. Restore to a new location by default, compare, then merge. Add this rule to the runbook.Teams messages are missing from the backup. Native Teams backup is not released. Use a vendor that lists Teams support or document the compensating control.The audit log looks empty. Ingestion may be off, may have been enabled recently, or your licence retention may be short. Run the check from Step 6 and wait for propagation.S3 bucket creation fails. Bucket names are global, so pick a unique one. Object Lock must be enabled at creation, so you cannot add it later to a plain bucket.A third-party tool fails to authenticate after a change. Some vendors are retiring Exchange Web Services calls in favour of Graph, as MSP360 did in version 4.11. Update the agent before you chase tenant settings.

Advanced Tips for Larger Practices and Groups

Use separate tenants or at least separate Azure subscriptions for production and backup billing, so a compromise of one does not expose the other. Put the guard script’s service principal on a certificate with a short lifetime, rotate it quarterly and restrict it to your management network. Export the drill CSV into a document store with write-once retention, then link each row to a ticket number for assessors.

If you run multiple clinics, give each site its own policy set and tag its restore targets, so one location can recover without waiting for another. Model your recovery order against clinical priority, not alphabetical order. Rehearse a day-one scenario where email is unavailable, and decide in advance how reception will reach patients by phone or SMS.

Watch the calendar too. Microsoft Ignite runs November 17 to 20, 2026, and backup announcements often land there. Treat any new Teams coverage as a trigger to rerun Steps 5, 10 and 12. The takeaway: a HIPAA Microsoft 365 backup is not a purchase, it is a loop of scope, protect, monitor and rehearse that you repeat every quarter.

FAQ: HIPAA Microsoft 365 Backup in Australia

Does Microsoft 365 Backup make my clinic HIPAA compliant?

No. Microsoft’s current documentation describes the service and its pricing but does not establish HIPAA or Australian healthcare compliance on its own. Compliance depends on your configuration, contracts, access controls and processes.

Does HIPAA even apply to an Australian healthcare provider?

HIPAA is a United States regulatory framework. It can apply if you handle data for US-regulated entities, and many providers use it as a technical benchmark. Your core obligations remain the Privacy Act, the Australian Privacy Principles and the Notifiable Data Breaches scheme.

Which Microsoft 365 workloads does the native backup cover?

Exchange Online, SharePoint Online and OneDrive for Business. Teams backup is not yet released and remains listed as coming in the future.

How much does Microsoft 365 Backup cost?

The list price is US$0.15 per GB per month of protected content, billed pay-as-you-go through an Azure subscription. Billing includes recycle-bin content and Exchange online archives. Customers using an independent software vendor’s product are billed by that vendor.

Do I still need a third-party backup?

Only if your risk assessment requires a copy outside the Microsoft trust domain, Teams coverage, or features the native service lacks. Many clinics run the native service for speed and add one independent immutable copy for resilience.

Where is my backup data stored?

It depends on the product and your region settings. Certification listings for third-party apps show multiple regions, including Australia and others. Get the storage and processing regions in your contract and verify them in the console.

What should I do first if ransomware hits?

Isolate accounts, preserve logs, freeze backup changes, and identify the last clean restore point. Then restore in stages and start your NDB assessment. The runbook in Step 13 lays out the order.

How often should I test restores?

Quarterly at minimum, and after any major tenant change. The guard script in this tutorial flags a missing passing drill after 90 days, which you can tighten in config.json.More on this topicCloud ComputingCybersecuritySoftwareAI & Machine LearningTrending nowTemporal Workers Hit Cloud Run at Almost 40% Less [2026]Google’s Agent Compute Engine: 1 microVM Per Agent [2026]DTA Releases Cloud Guidance Backing FinOps, 5 Rules [2026]Best Free Office Suite 2026: Free Office Software AlternativesHugging Face AI Hack Timeline: 4 Key Dates [2026]
Nadia DuboisAI & Innovation EditorNadia Dubois is the AI & Innovation Editor at Tech Insider, where she tracks the rapid evolution of artificial intelligence, from foundation models to real-world enterprise deployment. She previously covered AI and startups for La Tribune and contributed to MIT Technology Review’s European coverage. Nadia specializes in generative AI, AI regulation, and the intersection of technology and European industrial policy. She holds a dual degree in Computational Linguistics and Journalism from Sciences Po Paris.View all articles

Click Here For The Original Source.

——————————————————–

............