Between August 8 and August 18, 2025, attackers used stolen OAuth tokens from Salesloft’s Drift integration to retrieve data from more than 700 Salesforce environments and some connected Google Workspace inboxes. The vendor compromise had begun through its GitHub account months earlier. Cloudflare, Palo Alto Networks, and Zscaler were among the companies that confirmed exposure. The tokens let the attackers reach customer platforms through an already authorized integration.
Connecting a SaaS product to Google Workspace, Salesforce, or Slack gives it credentials that can act inside customer accounts. Review the permissions, token handling, and tenant mapping before customers enable the feature. These are the areas a scoped assessment should cover.
Why an integration is a different risk category than your own application
An integration adds a connection between your application and a platform you do not control. A customer authorizes it with a token. If that token leaks or your application selects the wrong tenant’s credential, the effect can reach beyond your own data to connected customer accounts.
In the Drift incident, the vendor intrusion preceded the token abuse. Once attackers could obtain valid integration credentials, they could query customer systems without stealing each customer’s login. Include the systems storing and using tokens in the assessment, alongside your application’s sign-in flow.
OAuth permissions and the scopes your integration requests
Compare the consent screen with what each feature needs. OWASP’s OAuth 2.0 guidance recommends limiting token privileges, resources, actions, and audience. A calendar-availability feature should not need broad mailbox access. Record why each requested scope is necessary.
Revisit that map when features change. A permission added for one release may remain after the feature is removed, or a broad grant may be used to avoid a more specific authorization flow. Match every current scope to an active feature and investigate unexplained permissions.
The authorization flow itself: PKCE, redirect URIs, and a deprecated grant type still showing up in the wild
Review the authorization flow against current provider and OAuth guidance. Use the Authorization Code Grant with PKCE where required for the client. Check for legacy Implicit Grant implementations and plan their replacement. PKCE helps bind redemption of an authorization code to the client that began the request.
Validate redirect URIs against the approved callback addresses. Check for wildcards and user-controlled parameters that can move the response to an attacker’s destination. Also verify how the flow binds the authorization response to the initiating user session, including PKCE, state, and any OpenID Connect nonce handling.
Token storage: the part of the system that turns a feature into a liability
Inspect where access and refresh tokens are stored and which services can read them. Verify encryption and access controls, then test error paths for leaks into logs, traces, or debug output. Review the provider’s refresh-token protections, including sender constraints or rotation where applicable. Keep live token values out of the assessment evidence.
A stolen token may still be accepted as valid by the provider. Test what happens when a customer disconnects: which grants are revoked, which local credentials are removed, and whether an earlier token still works. Deleting a database row does not itself revoke a token at the issuer.
Account isolation: the bug that crosses from your database into someone else's platform
Check how your application maps each customer to its external credentials and cached data. OAuth can work correctly while your application returns Company A’s records to Company B. OWASP’s multi-tenant guidance addresses this separate application-level problem.
Use two test tenants to check API requests, background sync jobs, caches, and webhooks. Verify that a client-supplied identifier cannot select another tenant’s token or data. Trace the authenticated tenant through the database and API client instead of stopping at the login check.
Webhooks: the third-party access risk running in the other direction
Include inbound event endpoints as well as outbound API calls. Webhooks are publicly reachable inputs to your application and can trigger lookups, updates, or background jobs.
Follow each provider’s documented webhook authentication and signature procedure before trusting the payload. Test stale or repeated deliveries, and verify that duplicates do not repeat a business action. Check payload fields used in queries, URLs, or downstream jobs for the same input-handling problems as other endpoints.
Why your customers' security teams will ask about this before enabling it
Customer security teams may ask for evidence before approving CRM or mailbox access. Keep a dated report that identifies the tested integration, scopes, account-isolation cases, and remediation status. It gives the reviewer something specific to assess and reduces follow-up questions about what a general compliance statement covers.
What a scoped assessment looks like before launch
Start with the OAuth callback, redirect handling, webhook endpoints, and requested scopes. Agree what configuration and test-account access the assessor needs to validate each one.
Add authenticated tests with at least two simulated tenants. Attempt the agreed cross-account requests, inspect token handling on error paths, and test denied or replayed webhook requests. Keep the resulting records and provider decisions with each finding.
Repeat relevant cases when adding scopes, webhook subscriptions, or customer roles. Include provider requirement changes in that review. Ceron offers an initial assessment, deeper authenticated testing, and recurring reviews on a no-findings, no-fee basis within the agreed terms. Choose the schedule around how often the integration changes.
A pre-launch checklist worth working through line by line
Before launch, account for every requested scope and validate the authorization flow and callback addresses. Check token storage, logging, renewal, and revocation. Verify tenant selection with separate test accounts. Confirm that webhook authentication, duplicate handling, and replay controls are enforced. Assign owners to unresolved items and specify which changes require a retest.
Review the integration before enabling customer access
Customers granting an integration access are relying on your application to protect that credential and use it for the agreed purpose. Test the token lifecycle and the tenant relationships before enabling customer accounts, then revisit them as the feature grows.