Guide, sources citedLast checked 16 October 2026 with AI-assisted research. Editor: Akshay Dubey.Corrections logEditorial policy
On this page
  1. Summary
  2. Timeline
  3. Attack chain
  4. NHI root cause
  5. Own disclosures
  6. How customers found out
  7. Impact assessment
  8. Mitigation playbook
  9. Lessons that transfer
  10. Primary sources
  11. FAQ

TL;DR

  • Drift (a Salesloft-owned SaaS) was compromised, and the OAuth tokens it held for customers' Salesforce integrations were stolen.
  • The attacker used those tokens to access Salesforce data at downstream customer organisations between 8 and 18 August 2025 (Google Threat Intelligence Group). Google told reporters that more than 700 organisations were potentially affected.
  • Our mapping to the OWASP NHI Top 10 (an interpretation, not a finding by Google, Salesloft or OWASP): NHI3, vulnerable third-party NHI, combined with NHI5 (broad OAuth scopes) and NHI7 (long-lived refresh tokens).
  • Salesloft and Salesforce revoked the Drift tokens on 20 August 2025 and notified customers. Google Threat Intelligence Group published its analysis on 26 August.
  • The mitigation playbook: revoke all Drift OAuth tokens, audit Salesforce data-access logs for the attack window, narrow OAuth scopes on re-install, and institutionalise OAuth-integration review cycles.

Summary

The Salesloft/Drift incident is, in our view, a clear large-scale example of OAuth supply-chain risk: when an upstream SaaS vendor holds OAuth refresh tokens granting broad access to downstream customers' other SaaS environments, compromise of that vendor cascades into every customer that authorised the integration. The population of impact was determined not by the attacker's choices but by who had ever installed the Drift→Salesforce integration.

The event was significant not because the primitive (OAuth token theft) was new, but because of its reach. One upstream compromise exposed the Salesforce data of every customer whose tokens Drift held, and Google put the number potentially affected at more than 700.

Timeline

Dates are from the sources named in each row.

DateEvent
Mar to Jun 2025The actor had access to a Salesloft GitHub account, according to the investigation Salesloft published with Mandiant (BleepingComputer report). It later reached Drift's AWS environment and obtained OAuth tokens for Drift customers' integrations.
8 to 18 Aug 2025The actor used the stolen tokens to query and export data from customers' Salesforce instances (Google Threat Intelligence Group). Cloudflare dates the activity in its own tenant to 9 to 17 August.
20 Aug 2025Salesloft, working with Salesforce, revoked all active access and refresh tokens for the Drift application. Salesforce removed Drift from AppExchange (Google Threat Intelligence Group).
23 Aug 2025Cloudflare was notified by Salesforce and Salesloft (Cloudflare).
26 Aug 2025Google Threat Intelligence Group published its analysis and tracked the actor as UNC6395.
28 Aug 2025Google widened its advice to all Drift integrations after finding that tokens for the Drift Email integration had been used on 9 August to read mail in a very small number of Google Workspace accounts.
2 Sep 2025Cloudflare published its account. Other affected companies made their own disclosures around the same time.

Attack chain

  1. Initial access at Salesloft. Per the investigation Salesloft published with Mandiant, the actor had access to a Salesloft GitHub account from March to June 2025 and then reached Drift's AWS environment.
  2. Token theft. From that environment the actor obtained the OAuth tokens Drift held for its customers' integrations. These tokens are the non-human credentials at the centre of the incident.
  3. Downstream access. Using the tokens, the actor authenticated to customers' Salesforce instances as the Drift connected app.
  4. Export and credential hunting. Google observed queries against the Account, Opportunity, User and Case objects, and searches of the exported data for secrets such as AWS access keys, passwords and Snowflake tokens.
  5. Covering tracks. Google reports that the actor deleted its query jobs. The logs were not affected, so customers could still review what was taken.

NHI root cause analysis

Our mapping to the OWASP NHI Top 10 (an interpretation, not a finding by Google, Salesloft or OWASP):

  • NHI3, Vulnerable third-party NHI. In our reading, the most significant factor. Each downstream customer had authorised Drift to hold an OAuth refresh token with broad Salesforce access. When Drift was compromised, those tokens became attacker assets. None of the downstream customers had compromised their own environments.
  • NHI5, Overprivileged NHI. The data reachable was set by the scope each customer had granted the Drift app. Where that scope covered support cases, anything customers had pasted into cases was exposed with it. Whether a narrower scope would have served Drift's purpose is a question each customer has to answer for its own tenant.
  • NHI7, Long-lived secrets. A refresh token stays valid until it is revoked or expires under the policy set for the connected app. The tokens taken here still worked when the actor used them.

Below these three sits a governance question we cannot answer from public sources: how many customers kept an inventory of their connected apps and reviewed the scopes they held. An integration that is installed once and never reviewed is the condition this incident exploits.

Organisations that published their own disclosure (selected)

Each organisation below said in its own statement that some of its Salesforce data was accessed through the compromised Drift integration:

Google is not on this list. Google Threat Intelligence Group said that tokens for the Drift Email integration were used to access a very small number of Google Workspace accounts, and that there was no compromise of Google Workspace or Alphabet itself, as reported by SecurityWeek.

The list is non-exhaustive. Google said more than 700 organisations were potentially affected, as reported by CyberScoop on 28 August 2025. That is an estimate of potential exposure, not a count of confirmed breaches.

How customers found out

In the public account we reviewed (Cloudflare), the customer learned of it through notification, not through its own detection. Cloudflare writes that Salesforce and Salesloft notified it on 23 August 2025, after the activity in its tenant had ended and three days after the tokens were revoked. Its review then found that the actor had exported the text of support cases, and it rotated 104 Cloudflare API tokens found in that data.

What this suggests for your own monitoring, as our reading and not a finding of any of the sources:

  • Keep a baseline of API usage for each connected app, so that a bulk export by an app that normally reads a few records stands out.
  • Alert on a large deviation from that baseline.
  • Keep the connected app authentication and API logs long enough to investigate after a late notification.
  • Know in advance how to revoke one app's tokens.

Impact assessment

  • Data exposure: customer CRM data, account details, contact information, case history, custom fields. Google reported that the actor searched the exported data for secrets such as AWS access keys, passwords and Snowflake tokens.
  • Operational disruption: the Drift integration was disabled for every customer when the tokens were revoked on 20 August 2025.
  • Secondary exposures: in cases where CRM records contained references to other credentials (API keys in support cases, passwords in customer communications), the exposure extended beyond the Salesforce data itself.

Mitigation playbook, what to do if you were impacted

  1. Confirm the Drift tokens are revoked. Salesloft and Salesforce revoked them on 20 August 2025. Check the connected apps list in your own Salesforce org.
  2. Pull Salesforce login and API-call history for the Drift OAuth app for the full attack window. Look for enumeration patterns, queries outside operational hours, IP addresses unusual for Drift.
  3. Identify data that was accessible. Map the Drift OAuth scope to specific Salesforce objects and fields. Any field reachable via those scopes is candidate-exposed.
  4. Audit CRM content for secondary credentials. Search case notes, custom text fields, and attachments for password/token/key patterns. Rotate any credentials found.
  5. Notify appropriately. GDPR Article 33/34, state AG notifications, customer notifications as applicable. Qualified counsel decisions.
  6. Review all other OAuth integrations. Put each one through the same cycle: inventory, scope review, rotation cadence and revocation of unused apps. Our SaaS integration governance page sets out that cycle.

Lessons that transfer to every organisation

  • OAuth apps are NHIs. Treat them with the same rigour as service accounts: inventory, a named owner, scope review and rotation. See OAuth for non-human identities.
  • OAuth scopes should be minimum-viable. Do not grant "read all" when "read a specific object" would suffice.
  • Supplier OAuth-token hygiene is your problem. You cannot assume your SaaS suppliers rotate their stored refresh tokens. Contractually, you can push for improvements; operationally, you can rotate on your schedule.
  • SSPM tooling can help. SaaS security posture management tools can provide a cross-SaaS OAuth inventory; see the identity threat detection and SaaS posture category.
  • Do not count on detecting it yourself. In the public account we reviewed, the customer was told by the vendors. A usage baseline for each connected app is what would let you see a bulk export without waiting for that call.
  • Third-party risk duties may reach OAuth integrations. NIS2 (law) and DORA (law) set third-party ICT risk duties. Reading OAuth integrations as covered by them is our interpretation.

Would your controls have changed the outcome? The breach readiness score asks you to rate your own controls against seven publicly reported incidents, including this one, and lists the scenarios to revisit. No sign-up is needed.

Other case studies: Codecov Bash Uploader compromise (2021), Microsoft Storm-0558 (2023), Okta support system breach (2023), Shai-Hulud 2.0 npm worm (2025), Snowflake and UNC5537 (2024).

Primary sources

Sources are listed below. Where our figure differs from an earlier published number, the entry says why.

Updates

If you spot a factual error or have additional verified information, email hello@nhigovernance.com.

Frequently asked questions

What was the Salesloft Drift OAuth breach?

OAuth tokens associated with the Drift integration were compromised and used to access data in connected systems belonging to downstream organisations. The integration itself was legitimate and had been authorised by each customer; the compromise occurred at the integration provider. In our view, it is a clear recent example of third-party non-human identity risk, where the customer's own controls were not the failure point.

Why do OAuth integrations create non-human identity risk?

Because they create a durable, broadly scoped credential held outside your organisation, usually granted once by a business user and never reviewed. The grant has no expiry by default, no named owner in your inventory, and permissions that reflect what the vendor asked for rather than what the integration needs. When the vendor is compromised, that grant is the attacker's access path and you have no visibility into its use.

How do you find third-party OAuth grants in your environment?

Through the admin consoles of the SaaS platforms themselves rather than through your identity provider. Each major platform exposes a list of authorised applications with their granted scopes. This is rarely in a security team's inventory because the grants are created by business users through consent screens rather than by IT through provisioning.

What is the right control for third-party integrations?

Restrict who can authorise them, review granted scopes against actual need rather than vendor request, register each as a non-human identity with an owner and review date, and rehearse revocation so that removing an integration under pressure is a known procedure rather than an improvisation. Scope reduction is the highest-value control, because it bounds the blast radius of a compromise you cannot prevent.

Which organisations were affected by the Salesloft Drift breach?

Google estimated that more than 700 organisations were potentially affected, which is an estimate of exposure rather than a count of confirmed breaches. Organisations that published their own disclosures include Cloudflare, Zscaler, Palo Alto Networks, Tanium, Cato Networks and Qualys. Google said tokens for the Drift Email integration were also used to access a very small number of Google Workspace accounts.

From the publisher

HumanAudit, which publishes this site, is preparing the NHI Governance pack: an NHI policy, ownership register, lifecycle procedures and one control set mapped, as our interpretation, to PCI DSS v4.0.1, ISO/IEC 27001:2022, NIS2 and DORA. Planned for March 2027; not yet on sale, and no price has been published.

Monthly change log

Know when the NHI landscape changes.

One email a month from this reference: new deals, standards status changes and corrections. No sponsored content. Unsubscribe from any email.

We first email you a link to confirm. Privacy policy.