When an employee connects an AI tool to Google Workspace, Microsoft 365, Salesforce, GitHub or Slack, the tool receives a token with its own scopes and lifetime. It does not need a password, it survives password changes and it is rarely reviewed again.
Two well documented breaches followed this route. In the Salesloft Drift breach, stolen OAuth tokens gave attackers Salesforce data within the grants' permissions, and the exports contained further credentials. In the Vercel breach, a token granted to an AI analytics tool led to a Google Workspace account, internal systems and customer secrets. Every step used access that had been granted on purpose.
A grant, a user account and a shared document. Each step is legitimate. Together they are a path Elmoz can trace and rank.
Short answers on OAuth grants to AI tools and how to review them.
They can be. An OAuth grant is standing access that does not need a password and usually does not trigger MFA. If the vendor behind the app is breached, the attacker can use the grant within its scopes.
In Google Workspace use the third-party app access page in the Admin Console, in Microsoft Entra the enterprise applications and user consents. Check scopes, owners and last use, and revoke apps nobody uses.
It connects each grant to what it can reach through the user account and further systems, so you can see which grants would lead to sensitive data if the vendor were compromised.
Every use case ends in the same place: a path from an identity to sensitive data.
Book a demo and we walk through the access paths in an environment like yours.