The threat actor tracked as UNC6395 stole OAuth and refresh tokens for Salesloft's Drift AI chat agent, then impersonated the trusted integration to query the Salesforce instances of more than 700 organisations between August 8 and 18, 2025. They walked out with AWS keys and Snowflake tokens. On August 20, Salesloft and Salesforce revoked every Drift token and pulled the app from the AppExchange.
Impersonation, but at the API layer
Deepfakes impersonate faces. This attack impersonated a trusted software integration. Same playbook, different layer. The Drift chatbot was a legitimate, widely installed connector, which is exactly why its tokens were valuable. Nobody double-checks the familiar app. The attackers did not break encryption or find a zero-day. They just wore a trusted uniform and walked through the front door.
OAuth tokens are bearer credentials: whoever holds them is authenticated, no questions asked. Once the tokens were compromised, every connected Salesforce instance treated the attacker as Drift. This is a supply-chain attack wearing an AI costume. You did not choose Drift's security posture; you inherited it the day someone on your team installed the integration. Third-party AI tools are now part of your attack surface whether you audited them or not. The blast radius, 700-plus customers, shows how much trust enterprises place in third-party AI integrations without verifying what is on the other end.
Deepfake detection api integration is about verifying media, but the principle generalises: never trust, always verify. A deepfake detection api checks that a face or voice is genuine before you act on it. Apply the same suspicion to every integration, token and agent touching your data. Trust is the vulnerability.
Incident facts
- Date: 2025-08-08
- Company / actor: Salesloft (Drift)
- Harm type: privacy breach
- Severity: high
- Source: Google Threat Intelligence Group