Claude Code Caught Using Dark Patterns to Hijack Your Google Account via MCP

Developers playing with the new command-line tool Claude Code were greeted by a bright orange, high-alert warning stating that authentication for three MCP servers was urgently required. For any sane programmer, a flashing warning usually means a critical dependency is broken or a security protocol has failed. In reality, this scary system prompt was just a glorified sales pitch trying to trick them into linking their personal Google Accounts.

Once users dug past the panic-inducing interface, they discovered these mysterious "required" servers were actually connectors for Gmail, Google Calendar, and Google Drive. It turns out Anthropic decided that the best way to roll out their new beta features was to dress them up as critical system errors. Because nothing says "trustworthy AI partner" quite like using the same interface design principles as a sketchy software installer from 2004.

According to Anthropic's official documentation, this integration is designed to let the AI browse through inbox archives and calendar schedules to answer user queries. The company was quick to assure everyone that Claude cannot actually draft, send, or edit emails, nor can it sneak a peek at inline images. The feature is currently in beta and is restricted to paying users on Pro, Max, Team, and Enterprise tiers, requiring workspace administrators to manually toggle it on before the AI can start sniffing around corporate emails.

The irony of an AI safety startup resorting to corporate growth-hacking tactics is peak Silicon Valley. It seems even the most righteous tech giants cannot resist the urge to peek into private emails under the guise of an urgent system upgrade. The line between a helpful assistant and a pushy adware installer is officially gone.

Source: Anthropic