A traditional vendor relationship has a contract, a procurement record and a named owner. A SaaS relationship often has none of those. It starts as a free trial, becomes a team subscription on a corporate card, and picks up data and permissions one connector at a time. By the time security hears about it, the vendor already holds production data. That is the first thing that is different about SaaS, and it means the vendor inventory is the weakest control in the program, not a solved starting point.
The second difference is where the risk sits. With a hosted-software vendor the vendor runs the stack, but you configure the tenant: who can sign in, whether single sign-on and MFA are enforced, which integrations are allowed, how long sessions and tokens live, and whether logs leave the platform. Responsibility for the outcome is shared, and a vendor's audit report speaks only to its half. The UK NCSC's guidance on using cloud services securely is a good plain-language statement of that split.
Two incidents that show where SaaS risk actually lives
Neither started with a breach of the core platform that held the data (Snowflake or Salesforce), which is why they are useful teaching cases.
- Snowflake customer instances, 2024. Mandiant's investigation of the campaign it tracks as UNC5537 reported that it found no evidence that access stemmed from a breach of Snowflake's own enterprise environment. Every incident it responded to was traced to compromised customer credentials, in the case it first described credentials previously stolen by infostealer malware. Mandiant pointed to customer-side controls: the affected accounts did not have MFA enabled, stolen credentials had not been rotated, and network allow lists were not in place.
- Salesloft Drift, 2025. Google's Threat Intelligence Group described a campaign (UNC6395) that, from as early as 8 August through at least 18 August 2025, used compromised OAuth tokens tied to the Drift application to export data from many customers' Salesforce instances, then searched that data for secrets such as AWS access keys and Snowflake-related tokens. The issue did not stem from a vulnerability in the core Salesforce platform. On 28 August GTIG widened its advice: tokens for other Drift integrations were also compromised, including the Drift Email integration, which was used to access email in a very small number of Google Workspace accounts, and customers were told to treat all tokens stored in or connected to Drift as potentially compromised. Salesloft later reported, from Mandiant's investigation, that the attacker had access to Salesloft's GitHub account from March to June 2025 and then took the OAuth tokens from Drift's AWS environment.
The Drift case is the one to put in front of a risk committee. The vendor that was breached was a connector. The customer's real exposure was the standing access that connector had been granted to the system of record, and the stolen data then fed a second round of compromise through the secrets it contained.
Why a questionnaire does not cover this
A security questionnaire or a SOC 2 report describes how the vendor runs its service. It will not tell you whether your tenant enforces MFA, which of your employees granted a third-party app read access to the mailbox, or how many connected apps still hold tokens from a pilot that ended last year. A SOC 2 report makes this explicit through its complementary user entity controls: the controls the vendor assumes you operate. See SOC 2 in vendor assessment for how to read them. In SaaS those controls are often most of the story.
A program that fits SaaS
- Discover, because the inventory is incomplete. Start from what is observable: the identity provider's application catalog, the list of OAuth grants and connected apps in your major platforms, expense and card data, and DNS or browser telemetry where you have it. Treat every discovered app as a candidate vendor record, then tier it. Expect the first pass to surface vendors that procurement did not know about.
- Tier by data and privilege, not by spend. A free tool with a read-write token into the CRM can matter more than a six-figure platform with no integrations. Add integration privilege and data sensitivity to the tiering criteria in vendor tiering. NIST SP 800-161 Rev. 1 (Update 1) likewise recommends prioritizing suppliers by criticality; see the publication.
- Inventory integrations as first-class assets. For each connector record the owner, the data it can reach, the scopes granted and the token lifetime. OAuth 2.0 (RFC 6749) lets a client request a limited scope and lets the authorization server grant less than was requested; in practice grants are often broader than the use case needs and are rarely reviewed again.
- Assess your tenant, not only the vendor. For high-tier SaaS, check SSO enforcement, MFA, admin role assignment, session and token lifetime, audit-log export, and which integrations are permitted. This is the customer-side counterpart to the vendor's evidence package.
- Write the exit into the contract. Ask for sub-processor disclosure and change notice, a breach-notification window, access to audit logs, a data export format and a deletion commitment with a date. The expectation that an external service provider meets your security requirements is a standard control: see NIST SP 800-53 Rev. 5, control SA-9 on external system services.
- Watch for the vendor's incidents, not just its annual report. In the Drift timeline, Salesloft and Salesforce revoked the Drift tokens on 20 August, about twelve days after the earliest data theft GTIG observed in customer Salesforce instances, and months after the attacker first got into Salesloft's environment. You cannot shorten the time before a vendor discloses, but subscribing to vendor status and security advisories, and having a runbook that rotates tokens for a named vendor, shortens the time between its disclosure and your response. See continuous vendor monitoring.
- Offboard the grants, not just the subscription. Cancelling a plan does not revoke an OAuth token or delete the data. Offboarding should revoke the integration, remove the user accounts and confirm deletion, as set out in vendor offboarding.
Integrations also create chains. A tool you approved connects to a platform you approved through a vendor you never heard of, which is the same structure as fourth-party risk, only created by your own employees.
Where this sits among TPRM platforms
SaaS sprawl is partly a discovery and tenant-configuration problem and partly a classic vendor-assurance problem, and no single tool class covers both. Platforms such as OneTrust and ProcessUnity are strong at intake, tiering, questionnaires and contract tracking once a vendor is known. Security-rating providers such as BitSight and SecurityScorecard add externally observed signal about the vendor, but an outside-in view cannot see your tenant settings or your OAuth grants. SaaS security posture tools, a separate category, cover the tenant side. Rescana's focus is the vendor side of that picture: continuous monitoring and automated evidence collection, so that an assessment of a SaaS vendor reflects current signal rather than the questionnaire answers from the day it was onboarded. Most programs will combine more than one of these.
Frequently asked questions
How is SaaS vendor risk different from other third-party risk?
With SaaS the vendor operates the infrastructure but the customer configures the tenant, so responsibility is shared. A vendor's SOC 2 report or questionnaire covers how the vendor runs its service, not whether your tenant enforces MFA, restricts integrations or exports logs. SaaS vendors are also adopted by individual teams without procurement involvement, so the vendor inventory is usually incomplete, and each integration adds standing access into other systems.
What is SaaS sprawl and why does it matter for third-party risk management?
SaaS sprawl is the accumulation of cloud applications adopted by teams outside central procurement and security review. It matters because every unreviewed app is an unassessed third party that may hold company data and hold OAuth tokens into core systems. A TPRM program that only covers vendors in the procurement system is blind to a large part of its real exposure, so discovery from identity-provider catalogs, OAuth grant lists and expense data has to feed the vendor inventory.
What did the Snowflake and Salesloft Drift incidents show about SaaS third-party risk?
Mandiant's investigation of the 2024 Snowflake customer thefts found no evidence of a breach of Snowflake's own environment and traced the incidents it handled to compromised customer credentials, which points to tenant identity controls. In the 2025 Salesloft Drift campaign, Google's Threat Intelligence Group reported that stolen OAuth tokens associated with Drift were used to export data from customers' Salesforce instances and then mine it for credentials. Together they show that the exposure often lies in customer-side identity settings and in the standing access granted to connectors.
How should a TPRM program handle OAuth integrations and connected apps?
Treat each integration as an asset with an owner, a record of the data it can reach, the scopes granted and a token lifetime. Review grants on a schedule, remove the ones tied to ended pilots or departed employees, request the narrowest scope that works, and have a runbook that revokes and rotates tokens for a named vendor when it reports an incident. Revoking the integration should also be part of vendor offboarding, because cancelling a subscription does not by itself invalidate a token.
Does a SaaS vendor's SOC 2 report cover my own tenant configuration?
No. A SOC 2 report covers the vendor's controls over its service. Controls the vendor expects the customer to run, such as user access reviews, MFA enforcement and the configuration of integrations, appear as complementary user entity controls, and the customer is responsible for operating and evidencing them.