Detailed Analysis
Claude's Gmail connector is triggering a "Google hasn't verified this app" warning screen during OAuth authorization, along with the unintended exposure of an Anthropic developer's email address in the authorization flow. The issue surfaced publicly when a user attempting to reconnect the Gmail integration in Claude encountered the warning and posted about it on Reddit. The warning itself is a standard Google security notice displayed whenever an OAuth application has not completed Google's formal verification process, meaning it remains in a testing or unverified state. Users can bypass the warning by clicking "Advanced" and then proceeding to the app, though the screen is designed to discourage casual users from authorizing potentially untrusted third-party access to their Google accounts.
The technical explanation for this behavior lies in how Google manages OAuth application states. Applications operating in testing mode — as Anthropic's Gmail connector appears to be — are subject to several restrictions beyond the warning screen, most notably that OAuth refresh tokens issued under this mode expire every seven days, forcing users to re-authenticate on a regular basis. This has practical implications for Claude users who rely on persistent Gmail integrations for automated workflows such as inbox triage, receipt extraction, or scheduled email summaries. For the connector to function with longer-lived tokens and without the warning, the underlying Google Cloud OAuth application would need to pass Google's full verification review, a process that involves demonstrating compliance with Google's API Services User Data Policy.
The exposure of a developer email address during the authorization screen is a separate and more sensitive concern. Google's OAuth consent screen surfaces the identity of the developer or project owner associated with an unverified app, which in this case appears to be an internal Anthropic account rather than a sanitized organizational identity. This is not a security vulnerability per se, but it reflects a gap in how the OAuth application was configured for external-facing deployment — production-grade integrations typically use a generic organizational identity rather than an individual developer's credentials as the listed contact.
The broader context here is that Anthropic has been rapidly expanding Claude's integrations with external services, including Gmail, Google Calendar, and Google Drive, as part of its push toward agentic AI workflows. The Gmail connector specifically is a component of the Model Context Protocol (MCP) tooling that allows Claude to read, compose, and manage email on behalf of users. The speed of this expansion has outpaced the slower, more bureaucratic process of obtaining formal verification from platform gatekeepers like Google. This is a familiar tension in developer tooling: features ship in functional but technically unpolished states, and compliance formalities lag behind. For users, the practical workaround remains straightforward — authorize through the advanced bypass path — but the developer email exposure and the seven-day token expiration represent friction points that Anthropic will need to resolve as the Gmail connector moves toward general availability.
Read original article →