IT security intelligence. Since 2006.
TRUST-IT / Practical guide

OAuth and SaaS compromise: investigate the application, not only the account

An unexpected connected application, a vendor breach notice or an unexplained data export needs more than a password reset. Establish which identity had access, what the application could do, what activity is recorded and which controls actually ended that access.

Illustrative enterprise team reviewing connected-application access
About 9 min read

TRUST-IT · Reviewed

What this investigation should establish

For enterprise security teams, business owners, legal advisers and family-office operations leads investigating an actual connection or incident.

  • Identify the application and its access path before selecting response actions.
  • Separate permitted access, recorded operations and demonstrated data transfer.
  • Agree a report, verified containment checks and explicit evidence limitations.
Scope a cloud investigation

Why review this now? Microsoft’s July 2026 research describes abuse of trusted SaaS OAuth relationships; its September research examines device-code phishing. Canada’s May 2026 advisory also addresses identity-led enterprise SaaS compromise. These are distinct mechanisms that justify reviewing application access alongside user accounts. They do not show that your tenant has been compromised.

Primary sources: Microsoft — SaaS OAuth abuse, 13 July 2026 · Microsoft — Device-code phishing, 22 September 2026 · Canadian Centre for Cyber Security — Enterprise SaaS compromise, 1 May 2026

Recognise the business question behind a connected-app alert

A CRM integration, document connector or reporting service may hold authorised access to sensitive information. A suspicious application therefore raises several different questions: was the connection approved, were its credentials exposed, was its permitted access used, and did data leave the service? A broad permission is a potential exposure. It is not, by itself, evidence that every accessible record was copied.

Set the scope around the affected tenant, application identifiers, business systems, relevant period and decision the report must support. Record the vendor notice or first observed symptom. Include the integration owner and an authorised administrator, together with legal or privacy advisers where appropriate. For a family office, keep the permissions and authority of separate companies, advisers and personal accounts distinct.

Separate four access paths before choosing containment

A compromised user session, a deceptively approved device-code sign-in, a malicious application consent grant and a compromised application identity are different access paths. In device-code phishing, a person may authenticate at a legitimate provider page while approving a session initiated elsewhere. Consent abuse instead concerns the permissions granted to an application. Do not assume that evidence of one establishes the other.

Distinguish delegated permissions, exercised in a user context, from application permissions used without a signed-in user. Identify the tenant-specific application identity, permission grant, affected resource and relevant credential or session. The responsible administrator should select the supported controls for that path. Resetting one user password is not a substitute for reviewing an independently authorised integration.

Primary sources: Microsoft — App consent grant investigation · Microsoft — Device-code phishing, 22 September 2026

ACCESS & EVIDENCE MAP

Follow the evidence across five distinct boundaries

This is an investigation map, not proof that an attacker completed every step. Each boundary needs its own evidence.

  1. AuthorityWho approved which app, grant or connection?
  2. IdentityWhich user, app identity or session obtained access?
  3. ResourceWhich mailbox, CRM object or file was reachable?
  4. ActivityWhich read, query or export is actually recorded?
  5. ConclusionWhat can be established, and what remains unknown?

Contain active risk in parallel. Permission does not equal use; use does not automatically prove export or identify the human operator.

Preserve the access map and the available records

Build an evidence register before the investigation becomes a collection of screenshots. Record application and tenant identifiers, the permission set, grant type, approving identity, creation and modification times, credential metadata, business owner and connected resources. Identify sources held by the identity provider, SaaS platform and integration vendor. Preserve relevant configuration and logs where feasible while necessary containment proceeds.

Check the actual coverage of each source: earliest available event, enabled categories, export permissions, licence dependencies and missing intervals. Identity-provider sign-ins, SaaS activity and mailbox audit records are separate datasets. Newly enabling a log does not recreate past events. Retain original exports with source, query, time range, timezone, collector, collection time and integrity hashes; verify pagination, errors and completeness.

Primary sources: Microsoft — Entra log retention

Evidence questions for the investigation brief

On small screens, swipe across the table or focus it and use the arrow keys.

Suggested questions; the available evidence determines what can be answered.
QuestionRecords to seekInterpretation limit
Was the connection authorised?Application identity, grant history, permission changes, owner approval and vendor connection record.A familiar display name or verified publisher does not establish approval of this particular grant.
Which access path was used?Relevant user, application and session records, credential metadata and provider correlation identifiers.A successful sign-in alone does not establish subsequent data access.
What happened to the data?Resource activity, API operations, export records and associated object or destination identifiers where available.An export request without completion evidence may not prove a completed transfer; missing events leave a gap.
Did containment take effect?Timestamped changes, service-specific revocation results, authorised checks and monitoring exceptions.A directory change does not demonstrate that every separately managed application session ended.

Test the sequence from permission to recorded activity

Construct a timeline that connects the grant or credential change with authentication, API activity and resource operations where records allow. Prefer stable application, tenant, session and correlation identifiers over display names. Preserve original timestamps and document any normalisation. A token issuance event indicates an access opportunity; it does not automatically describe the files, messages or records subsequently accessed.

Compare the observed activity with the integration owner’s expected workflow. A large export might be a scheduled backup, an approved migration or unauthorised activity. Examine affected objects, operations, volume, timing and destinations where available. Shared cloud addresses, proxies and normal provider changes require context. An IP address alone neither identifies a person nor establishes intent.

Contain each confirmed access path and verify the result

Agree who can disable an application, remove a grant, rotate exposed credentials or revoke sessions, and which business processes may be interrupted. Use the provider’s supported incident procedure, record the changes and preserve useful evidence in parallel. Do not delay urgent authorised containment until every export is complete. Avoid indiscriminate deletion of applications or logs that destroys context without a clear containment benefit.

Verification must cover the affected services, not just the identity directory. A SaaS application can maintain its own session, and token or policy changes do not take effect uniformly. Record what was revoked, when, how the result was checked and which residual sessions or dependencies remain unresolved. Reconnect a business integration only after the approved permissions, credentials and operating owner have been reviewed.

Primary sources: Microsoft — Revoke user access in Microsoft Entra ID

Bring the vendor into a defined evidence request

Ask the vendor to identify the affected customer connection, incident window, credential types, relevant API operations and available records for your tenant. Request the basis for statements such as “no customer impact”, including telemetry coverage and known gaps. A general public notice may not answer whether your particular integration was used. Route evidence exchange through authorised contacts and an agreed secure channel.

Separate facts established in your environment from vendor statements and hypotheses awaiting confirmation. Use a dependency list to track downstream systems and their owners. If an integration can reach a repository, HR platform or document store, examine the authorised access boundary before assuming lateral access. Preserve the notice and subsequent clarifications as dated evidence, rather than treating the latest email as a complete incident history.

ILLUSTRATIVE EXAMPLE

An application permission and an export event do not tell the same story

Illustrative scenario, not a client case: a company receives a notice about a compromised CRM integration. Its register shows broad read permission. Tenant records then show an unexpected API query, but the available export log does not cover the whole incident window. The report can describe the grant and recorded query, investigate its purpose and identify the missing period. It cannot truthfully claim that every customer record was stolen—or that no records left the service. The next request to the vendor targets the missing export and connection evidence.

Commission a report that distinguishes exposure from demonstrated impact

The findings should separate granted capability, observed authentication, recorded resource access, evidenced export and any supported onward disclosure. Explain the evidence and limits at each step. Where records are absent, identify which question remains unanswered. Do not describe a clean endpoint scan or lack of suspicious user sign-ins as proof that a connected application was not abused.

Agree an executive chronology, technical evidence index, access-path map, scope limitations and remediation register with owners and verification criteria. Clarify which platform configuration changes and retests are included. Legal, regulatory and contractual decisions should be made with the appropriate advisers using the verified facts; a technical report does not independently determine every notification obligation.

Turn the incident into a narrower, reviewable access model

Review which connections remain necessary and who accepts their permissions. Make integration approval, credential handling, vendor changes and retirement part of the operating process. An AI assistant that connects to company documents also belongs in this register: review the connector’s access separately from the model’s answer quality. Test with harmless records and explicitly authorised accounts rather than exposing production information to demonstrate a control.

For an initial TRUST-IT enquiry, state the platform, approximate discovery time, whether activity appears ongoing, the known application or vendor and who can authorise work. Agree availability, evidence access, deliverables and fees before the engagement. Use the public form for a short description only; arrange a secure channel for logs, notices or confidential records. Never paste passwords, access tokens or recovery codes into the form or chatbot.

A scoped engagement with usable outputs

  • An approved scope identifying systems, entities, relevant period and reporting recipients.
  • A connected-application access map and evidence coverage register.
  • A referenced chronology with separate findings, interpretations and unresolved questions.
  • A containment and remediation record with owners, verification results and residual exceptions.

Have a suspicious connected application or vendor incident?

Describe the platform and concern. Availability, authority, evidence access and fees are agreed before work; confidential material belongs in an agreed secure channel.

Scope a cloud investigation

Further reading

Put the guidance to work

Investigations across Microsoft 365, Google Workspace and cloud environments, including business email compromise, access and data sharing.

Email & cloud investigations

What’s your next
technology challenge?

TRUST-IT / FIND YOUR NEXT STEP

How can we help?

Popular topics

Search the public TRUST-IT website.

    TRUST-IT · AI assistant
    TRUST-IT

    How can we help?

    Explore our services, ask a question or tell us what you need. You will be speaking with our AI assistant.

    When you continue, Ultimo-Bots loads the chat. Your name, email and conversation are recorded so TRUST-IT can respond, and our team receives enquiry notifications. Phone, company and role are optional.

    Please do not share passwords, evidence files or sensitive information.

    Read our privacy notice

    Chat runs through Ultimo-Bots. Your name, email and conversation are recorded to handle your enquiry. Privacy notice

    Prefer to speak with our team? Contact TRUST-IT