Guide Cybersecurity Compliance

Reducing Microsoft dependency: the case for change and a practical roadmap

Microsoft provides the operating system, identity, email, collaboration, security tooling and cloud infrastructure for many organisations. That integration can lower operating costs and simplify administration, but it also creates dependencies worth assessing. Security incidents, service outages, data sovereignty requirements and licensing terms are all reasons to look at concentration risk. This guide explains how to identify those dependencies and decide when to strengthen your existing estate, diversify selected services or plan a migration.

Ryland Deakin
Ryland Deakin
Lead Consultant, Cyvra
12 August 2026
12 min read
Updated 9 November 2026
Key takeaways
  • The US Cyber Safety Review Board called the Storm-0558 intrusion "preventable". Microsoft reports that the forged tokens were used against Outlook Web Access in Exchange Online and Outlook.com, affecting 22 organisations and more than 500 individuals.
  • US law, including the CLOUD Act, can require a US provider to disclose data in its possession, custody or control wherever it is stored. Disclosure needs valid legal process, and providers can challenge some demands. It is a risk to assess, not an automatic breach of UK GDPR.
  • Federating sign-in to another identity provider does not take Microsoft out of the Microsoft 365 trust chain. Entra ID still validates the federated token and issues access to Microsoft services.
  • Most organisations do not need to leave Microsoft. Map which business functions depend on it, test what happens when each part fails, and then decide whether to strengthen, add independent fallbacks or diversify selected workloads.

The security incidents

Two incidents shape most discussions of Microsoft risk. They failed in different ways, and the lessons differ.

Storm-0558, a group Microsoft attributes to China, obtained a Microsoft consumer account (MSA) signing key. Combined with a token validation flaw, the key let the group forge tokens that enterprise Exchange Online accepted. Microsoft's investigation states that the only services where it observed the actor using forged tokens were Outlook Web Access in Exchange Online and Outlook.com. Victims included US government officials, among them the Commerce Secretary. The intrusion was first detected by a customer, the US State Department, through its own audit logging.

CSRB finding, March 2024

The US Cyber Safety Review Board concluded that the intrusion was "preventable and should never have occurred". It counted 22 affected organisations and more than 500 individuals, criticised Microsoft's security culture and found that Microsoft had been slow to correct inaccurate public statements about how the key was obtained. The board also noted that the key, with the validation flaw, could in principle have given access to Exchange Online mailboxes far more widely than it was actually used.

That distinction matters. The evidence shows broad potential access to Exchange Online and confirmed misuse against a limited set of mailboxes. It does not show that SharePoint, Teams or Azure were accessed with the same technique. The lesson for architects is still serious: when a provider's identity or signing infrastructure is compromised, the impact depends on which services trust it, and controls inside your own tenant may not prevent it.

Midnight Blizzard (also known as NOBELIUM or APT29) is a separate case. In January 2024 Microsoft disclosed that the group had password-sprayed a legacy, non-production test tenant account without MFA, then used it to reach corporate email, including that of senior leadership. Microsoft later reported that the group had accessed some source code repositories and internal systems, and it contacted customers whose correspondence with Microsoft contained secrets such as credentials. Here the failure started in Microsoft's corporate environment rather than in the authentication service customers use. It shows that a provider's internal security can affect its customers even when their own tenants are well configured.

Customer controls still matter in both scenarios. Audit logging is how the State Department spotted Storm-0558, and rotating any secrets shared with a provider limits the damage from an incident like Midnight Blizzard.

Data sovereignty and the CLOUD Act

Under 18 U.S.C. §2713, added by the CLOUD Act in 2018, US authorities can require a provider subject to US jurisdiction to preserve or disclose data within its possession, custody or control, regardless of where the data is stored. A Microsoft data centre in Dublin or Amsterdam does not by itself remove that exposure.

The power has limits. A demand needs valid US legal process, such as a warrant. Under 18 U.S.C. §2703(h), a provider can ask a court to modify or quash certain demands that would risk breaching the law of a qualifying foreign government. Microsoft publishes the number of law enforcement requests it receives, including those for enterprise customer data. The UK also has a Data Access Agreement with the US, in force since October 2022, which governs some cross-border requests between the two countries.

For a UK organisation, this exposure does not automatically mean your use of Microsoft 365 breaches UK GDPR. It is one factor in your supplier risk assessment and in your international transfer analysis, alongside encryption and key custody, provider access controls and contractual terms. It helps to keep the questions separate:

RiskWhat to assess
International transfersData flows, parties, locations, lawful basis and transfer mechanism under UK GDPR
Foreign government accessProvider jurisdiction, legal process, safeguards and the provider's ability and commitment to challenge requests
ConfidentialityEncryption, who holds the keys, privileged provider access and access logging
Data residencyWhere data is stored and processed, documented exceptions and support access
ContractNotice of requests, transparency reporting, incident notification and subprocessors
On the record, June 2025

Anton Carniaux, director of public and legal affairs at Microsoft France, was asked by a French Senate committee whether he could guarantee that data held for French public bodies would never be passed to US authorities without French authorisation. He said he could not guarantee it, and added that, to his knowledge, it had not happened. The testimony shows the limits of the legal assurance a US provider can give. It does not show that such disclosures routinely take place.

Microsoft's EU Data Boundary commits to storing and processing defined categories of customer data, pseudonymised personal data and professional services data for covered services within the EU and EFTA. Microsoft's documentation lists exceptions, including some security operations, remote support, customer-initiated transfers, network transit and specific service functions. The UK sits outside the EU Data Boundary. UK customers should check which data and services fall under Microsoft's UK data residency commitments and which flows can still leave the UK, rather than assuming either complete residency or unrestricted transfer.

European governments have also scrutinised this area. Germany's Federal Office for Information Security (BSI) analysed Windows 10 telemetry in its SiSyPHuS study and published guidance on limiting it. The EU Data Act, applicable from 12 September 2025, gives cloud customers in the EU rights to switch providers and port their data. It does not apply in the UK, but it affects UK organisations with EU operations.

Concentration risk in a single-vendor estate

Microsoft concentrates several critical functions in many enterprise environments: the operating system, on-premises identity (Active Directory), cloud identity (Entra ID), email, collaboration, security tooling and cloud hosting. That integration has real benefits, including fewer integrations to secure and one administrative model. It also means a failure in a shared component can affect several business functions at once.

The useful question is which failure would stop which part of your business. A failure-domain view answers it better than a list of incidents:

Failure domainWhat could stop workingPossible mitigation
Entra ID authenticationSign-in to Microsoft services and to other applications that use Entra IDTested emergency access accounts and, where justified, an independent access path for critical systems
Exchange OnlineEmail and processes that depend on itAn independent emergency communication channel and documented manual procedures
Azure hostingApplications, databases and supporting servicesApplication-specific recovery, tested cross-region or cross-provider arrangements
SharePoint and OneDriveFile access and collaborationOffline copies of critical documents such as continuity plans and contact lists
Azure Front Door and DNSApplications that depend on that ingress and routingIndependent routing or failover where the application justifies it
Admin and security portalsAdministration and incident investigationAlternative evidence sources, recovery credentials and an emergency response procedure

Diversity helps only where the alternative is actually independent. A workload moved to AWS may still sign in through Entra ID. A backup chat tool may use the same single sign-on and fail during an identity outage. Two services may share a DNS provider, a network link or a privileged access system. A secondary site may exist but hold stale data or an untested failover procedure.

Windows Recall shows how new AI features can raise privacy questions on the endpoint. The version first announced in 2024 drew criticism after researchers found its snapshot database was stored unencrypted. Microsoft changed the design before release: Recall is opt-in, runs only on Copilot+ PCs, stores snapshots locally in encrypted form behind Windows Hello Enhanced Sign-in Security, and administrators can disable or remove it on managed devices. Organisations should still decide, against their privacy and retention policies, whether to allow it on corporate devices.

Service outages and operational resilience

Large shared platforms fail from time to time. On 9 October and again on 29 October 2025, Azure Front Door incidents disrupted access to Azure and Microsoft 365 services for many customers; Microsoft attributed the 29 October incident to a faulty configuration change. On 22 January 2026, a Microsoft 365 incident affecting North American infrastructure disrupted Outlook, Teams, SharePoint, OneDrive and admin services for affected users.

The scope of an outage depends on the region, the components involved and which services a tenant uses. Not every incident affects every customer. When you depend heavily on one platform, though, a single incident can stop email, chat, files and sign-in together. For each critical business function, work out what stops in each failure scenario and keep tested contingency procedures for it. Recovery depends on the alternatives you have built and tested in advance.

Licensing costs and lock-in

Microsoft spreads advanced identity, threat protection, audit and compliance features across plans and add-ons. Privileged Identity Management needs Entra ID P2, Audit (Premium) comes with E5 or an E5 Compliance add-on, and Defender for Endpoint Plan 2 is in E5 or the E5 Security add-on. An organisation on Microsoft 365 E3 can often add these through add-ons or standalone licences instead of moving everyone to E5. Compare full E5, E3 with add-ons and third-party products on technical requirements, eligibility, coverage and operating cost, not only the per-user price.

Moving Microsoft workloads to another cloud can be harder than keeping them on Azure. Licensing terms for running Windows Server, SQL Server and Microsoft 365 apps on other providers' infrastructure differ, and some Azure-specific services have no direct equivalent. The cost structure can favour staying with the same provider, which is one reason to plan exit options before you need them.

Microsoft Secure Score is a measure of configuration against Microsoft's recommendations, and some of its recommendations depend on licences. Many improvements, such as access governance, removing legacy authentication and acting on alerts, need configuration and process rather than new purchases.

How organisations are responding

The most visible responses come from the public sector. In 2026, France's interministerial digital directorate (DINUM) announced steps to reduce dependence on non-European suppliers, including moving its own workstations to Linux and asking ministries to draw up dependency reduction plans. How far that spreads will depend on each ministry. Germany's state of Schleswig-Holstein is moving around 30,000 workstations from Microsoft Office and Exchange to open-source alternatives, with Linux on the desktop as a later phase.

Commercial organisations more often reduce concentration selectively. They replace Microsoft where the risk or cost is highest and keep it where the switching cost outweighs the gain. A full exit can make sense for organisations with strict sovereignty requirements, open-source mandates or estates small enough to keep migration costs manageable.

How to assess concentration before deciding to migrate

Before choosing replacement products, establish whether your concentration is actually a problem and which part of it. These questions give you that picture:

AreaQuestion
Critical servicesWhich business operations depend on Microsoft services?
IdentityWhich applications become unreachable if Entra ID fails?
Data and portabilityCan critical files, mailboxes, workloads and configurations be exported or recovered?
RecoveryAre alternative access, communication and recovery procedures tested?
Legal exposureWhich entities, transfers and jurisdictions apply to your data?
Commercial exposureWhat are the renewal terms, licensing alternatives and exit costs?
Operational independenceWould the alternative avoid the same failure domains?
Security impactWould diversification improve or weaken your ability to manage access, monitoring and patching?

The answers can support quite different decisions. One organisation keeps all of Microsoft 365 but adds independent emergency communications and backup. Another moves selected workloads elsewhere. A third deliberately stays integrated because a multi-provider estate would add more risk than it removes. Compare the options before committing to one:

OptionPotential benefitMain limitation
Strengthen the existing Microsoft estateLower migration risk and faster security and resilience gainsPlatform dependency remains
Add independent recovery and emergency accessLess disruption during outagesDoes not remove dependence on the primary provider
Diversify critical workloads selectivelyReduces some shared failure modesAdds operational complexity and integration cost
Federate identity or use a separate IdPCan improve identity governance across many applicationsEntra ID remains in the path for Microsoft services
Replace selected Microsoft servicesReduces exposure to a specific product or providerNeeds compatibility, migration and operational testing
Extensive multi-cloud or non-Microsoft migrationCan meet strong sovereignty or concentration requirementsHigh cost, complexity and ongoing operational burden

A workable sequence is: map dependencies, identify the critical failure scenarios, assess legal and commercial exposure, compare the options above, pilot where the case is made, then test recovery and exit arrangements.

Reducing dependency in practice

Before any migration, identify technical dependencies, contracts, regulatory requirements, recurring costs, transition effort and the risk of disruption. Prioritise changes whose measurable benefit to resilience, security or cost outweighs the risk and investment. The products below are examples to evaluate against your own requirements, not drop-in replacements.

Recovery and emergency access first

For many organisations, the quickest gains come from independent fallbacks rather than new platforms: tested emergency access accounts, an out-of-band communication channel for incidents, offline copies of continuity plans and contact lists, and backups of Microsoft 365 data held outside the tenant. These reduce disruption during an outage or compromise without migrating anything.

Identity

A third-party identity provider such as Okta, JumpCloud or Ping Identity can give non-Microsoft applications a common sign-in layer and reduce some operational dependencies. When users sign in to Microsoft 365 through federation, Entra ID still validates the federated token and issues access to Microsoft services, so federation alone does not remove risks in Microsoft's identity infrastructure. Decide based on which applications, data and services depend on Entra ID today.

Infrastructure

Moving existing Azure workloads carries cost and risk, and some organisations have commercial commitments, integrations, automation and security processes built around Azure. Where a new workload has no strong reason to sit on Azure, assess other providers, including European ones such as OVHcloud or Hetzner, on availability, security, jurisdiction, portability, support, cost and their own third-party dependencies. A European headquarters can reduce some legal exposure, but the provider may still have US operations, subprocessors or software dependencies, so it does not guarantee full data sovereignty on its own.

Email and calendar

Email is often the first service organisations consider replacing. Moving away from Exchange Online affects retention, eDiscovery, data loss prevention, mobile device management, identity integration, meeting scheduling and regulatory record-keeping, so every migration needs data, identity and mail-flow testing. Google Workspace diversifies the supplier, but Google is also a US provider, so it does not address the jurisdiction concern described above. Proton Business offers end-to-end encryption between Proton users and zero-access encryption for stored mail; messages to external recipients are protected end-to-end only with password-protected email or PGP.

Productivity applications

Google Workspace, LibreOffice and ONLYOFFICE can cover documents, spreadsheets and presentations for some teams. Test file compatibility with your own documents, especially Excel workbooks with macros or VBA and files exchanged with partners who still use Microsoft Office.

Endpoints

Endpoint migration usually takes the longest. Developer and technical teams are often the easiest group to move to Linux if their tools run on it. For general staff, plan application compatibility testing, device management and support before a structured rollout. macOS is another option where device cost, management tooling and application compatibility fit.

File storage and collaboration

Nextcloud can cover part of the storage and collaboration functions of OneDrive and SharePoint, depending on your requirements. Slack, Mattermost, Google Meet and Zoom are alternatives to parts of Teams. If you keep Microsoft 365, harden it with Conditional Access, send logs to a SIEM you can reach independently, and set telemetry controls explicitly.

Whatever you decide, write an incident response plan that covers Microsoft services being unavailable or compromised at the provider level, enforce MFA on every account, and apply Zero Trust access controls so that critical systems do not depend on one platform being available.

Risk assessment

Understand where Microsoft sits in your risk profile

We map your Microsoft dependency, identify your highest-risk exposure points, and help you build a realistic reduction plan.

Ryland Deakin
About the author
Lead Consultant, Cyvra · CISM · CompTIA Security+ · MCP

Ryland has delivered cybersecurity, compliance, and IT management programmes for regulated organisations across the UK and the Netherlands for over 20 years. View full profile