← All articles
Exchange Online · Email Security · Technical Guide

Disabling Direct Send in Exchange Online without breaking business mail

Understand the control, inspect your dependencies, deploy it safely and manage the operational risks of blocking legacy application mail.

Historical investigation, not a breach attribution. The original scenario involved anonymous external SMTP traffic and an apparently internal sender. It was consistent with Direct Send-style spoofing, but did not establish account compromise or prove the exact cause. Examples are generalised and sanitised.

Microsoft announced general availability of Reject Direct Send on 16 November 2026 for worldwide Exchange Online tenants. This guide revisits a real-world diagnostic problem through Microsoft's current supported control.

Why an internal-looking email is not necessarily from an internal account

During an October 2026 email investigation, the original message headers showed an anonymous external SMTP hop into Microsoft 365 and failed sender authentication checks. A subsequent gateway hop showed different authentication results. This illustrates why investigators need to distinguish original ingress from later forwarding. The observations were consistent with Direct Send-style spoofing; they did not independently prove mailbox compromise, successful exploitation of a particular configuration, or a completed remediation.

Microsoft introduced Reject Direct Send in public preview in April 2025 and announced general availability for worldwide Exchange Online tenants on 16 November 2026. This article revisits the historical investigation in light of that announcement. GCC High and DoD were not supported at the announcement date.

What Direct Send does

Direct Send allows devices and applications to deliver messages to internal Exchange Online recipients using the tenant's Microsoft 365 mail endpoint over TCP 25 without authenticating to a mailbox. It is not a general-purpose outbound relay. Legacy printers and monitoring applications are common dependencies. Normal inbound filtering still applies.

Distinguish the SMTP envelope sender (P1 MAIL FROM) from the visible message sender (P2 From). RejectDirectSend evaluates P1. It rejects anonymous messages when that domain belongs to the tenant's accepted domains and no qualifying inbound connector identifies the source by sending IP or certificate. It does not independently block every visible From spoof or every attempt to send directly to the Microsoft endpoint. Accepted subdomains are covered when MatchSubDomains is enabled.

Identify legitimate usage before enforcement

In the Exchange admin center, navigate to Reports > Mail flow > Change Optics Report and review the DRS scenario. The report provides example traffic, not a guarantee that every rare or intermittent sender is represented. Review a representative business cycle, including month-end workloads and emergency notifications. If the report is unavailable, record that visibility limitation, review trace samples and ask application owners for corroboration.

Use read-only Exchange Online PowerShell to inventory the current configuration:

powershell
Connect-ExchangeOnline
Get-OrganizationConfig | Format-List RejectDirectSend
Get-AcceptedDomain | Select-Object Name,DomainName,DomainType,MatchSubDomains
Get-InboundConnector | Select-Object Name,Enabled,ConnectorType,
    SenderDomains,SenderIPAddresses,TlsSenderCertificateName,
    RestrictDomainsToIPAddresses,RestrictDomainsToCertificate,RequireTls

Also review SPF records, known relay hosts, printers, applications, monitoring alerts, third-party gateways and forwarding arrangements. SPF is a useful starting point, not a complete inventory or a substitute for connector attribution.

For each identified legitimate sender, record the business owner, source, envelope domain, frequency, recipient scope, existing connector, replacement option and verification evidence. Do not create broad IP allowlists or trust shared infrastructure without understanding who can send through it.

Choose an appropriate replacement

OptionAppropriate useKey dependency
SMTP submission using modern authenticationApplication capable of authenticatingSupported OAuth implementation and authorised mailbox
Microsoft 365 SMTP relayApproved relay/device infrastructureAppropriate inbound connector and supported hosting scenario
Partner inbound connectorKnown service sending into the organisationValidating TLS certificate (preferred) or tightly scoped sending IPs
High Volume EmailEligible internal application notificationsVerify current service support, authentication and limitations
Dedicated application email serviceWorkload cannot use supported Microsoft optionsSeparate security and delivery design

Use Microsoft guidance for each option. A connector should assert a known source, not create a general anti-spam bypass. Review Enhanced Filtering for Connectors when a third-party gateway sits upstream.

Verify connector attribution

Collect a known legitimate message ID from a controlled test and inspect trace details:

powershell
Get-MessageTraceV2 -MessageId '<actual-test-message-id>' |
    Get-MessageTraceDetailV2 | Format-List

Check the actual routing and attribution details against the intended connector. A successful mailbox delivery alone is not sufficient evidence that the expected connector was used.

Enable rejection in an approved change window

The administrator needs the Organization Configuration role. Capture the existing value and the approved sender inventory before proceeding.

powershell
Set-OrganizationConfig -RejectDirectSend $true
Get-OrganizationConfig | Format-List RejectDirectSend

Allow up to 30 minutes for propagation, as stated in the GA announcement. A matching rejected message receives SMTP error 550 5.7.68 TenantInboundAttribution. Coordinate verification with sending-system owners, since rejected SMTP transactions may not appear like successfully accepted messages in ordinary message trace results.

Test matrix and business impact

ScenarioExpected result
Approved connector-attributed applicationContinues through normal filtering
Authenticated user or supported submission flowContinues, subject to applicable controls
Approved legacy device using unauthenticated Direct Send without a matching connectorRejected if its envelope domain is accepted
Controlled negative test using an owned lab domainRejected when all control conditions match
External legitimate sender using a nonaccepted envelope domainNot rejected solely by this control; normal filtering applies
Message with misleading visible From but a different envelope domainNot blocked solely by Reject Direct Send; evaluate anti-spoofing controls
Message forwarded back from a service that preserves the original accepted envelope sender without SRSPotential rejection without appropriate attribution

Run negative tests only in a lab or explicitly authorised environment. Confirm scan-to-email, batch jobs, ticket notifications, mail gateway delivery, disaster-recovery alerts and external forwarding with business owners.

Rollback and ongoing monitoring

For existing tenants where changing the value back is supported, restore the recorded prior setting only if the agreed rollback criteria are met:

powershell
Set-OrganizationConfig -RejectDirectSend $false
Get-OrganizationConfig | Format-List RejectDirectSend

This reopens the unauthorised Direct Send path, so treat it as temporary and explicitly risk-accepted. Microsoft is rolling out mandatory rejection for new tenants; do not promise this rollback is always possible. Prefer remediating the specific legitimate sender via a supported method.

Investigation lessons

Review original trusted Received hops, Authentication-Results at the correct boundary, connector attribution, envelope sender and visible From independently. Later SPF or DMARC outcomes do not necessarily establish original sender legitimacy. Correlate with Exchange and Entra audit evidence before asserting account compromise. Keep SPF, DKIM, DMARC, impersonation protection and third-party gateway routing controls in place: Reject Direct Send is complementary.

References

First published 21 November 2026. This article discusses a historical, anonymised scenario. The procedures are intended for authorised review and testing; no specific production change or successful lab test is asserted.